SlideShare a Scribd company logo
1 of 8
Download to read offline
Bio-inspired, Learning and Intelligent systems for security
                                      Bio-inspired, learning    intelligent Systems for Security




      Artificial intelligence methodologies applicable to support the decision-making
                        capability on board Unmanned Aerial Vehicles

                                                    Isabella Panella
                               Thales UK, Aerospace Division, Manor Royal, Crawley, UK
                                         isabella.panella@uk.thalesgroup.com


                             Abstract                                  systems, as single methodologies cannot address
    The need for Unmanned Air Vehicles (UAVs) to operate               independently the complexity of the problem of machine
    autonomously and to manage their operation with                    autonomy.
    minimal intervention from the ground control station, in               The architecture specifies the relationship between the
    order to reduce the datalink utilization and maximize              various subsystems that are seen to be key enablers to the
    their exploitation in beyond line of sight (BLOS)                  autonomy of the vehicle combined with legacy systems
    operations, has been long recognized within industry and           and the artificial intelligence (AI) techniques that could be
    research institutes. Many artificial intelligence (AI)             used in order to implement them.
    techniques try to address the challenge of moving UAV                  The motivation behind this work is to provide a
    towards full autonomy. However, no single technique has            possible UAV functional architecture within which the AI
    been able to provide the required autonomy for                     methodologies can be embedded and the need for them to
    unmanned platforms. This paper presents a Unmanned                 complement each others weaknesses and interrelationship
    Air Systems (UAS) architecture within which the different          appreciated.
    AI methodologies applicable to each subsystem are                      In this paper, unmanned air platforms refer to both the
    presented.                                                         air vehicle and the payloads fitted on the air vehicle in
                                                                       order to provide the required capabilities for the given
                                                                       mission.
    1. Introduction                                                        The paper is organised as follows. Section 2 provides a
                                                                       definition of automatic versus autonomous. These terms
        It has long being recognized that the employment of            are often confused and misused and it is appropriate to be
    Unmanned Air Systems (UAS) could provide significant               clear about the difference between them. Section 3
    advantages in military applications for ‘dirty, dull, and          provides an outline of the major functional elements of the
    dangerous’ missions as well as for civil and commercial            platform and the payload is also provided.
    applications, such as search and rescue, border                        Section 4 reports some of the key mission level
    management, pipeline monitoring.                                   functionalities and capabilities that are required for the
        UAV performance can be improved by increasing the              systems to work autonomously. Section 5 identifies the AI
    autonomy on-board of the platforms in order to reduce the          fields and techniques that can be applicable to solve some
    workload placed on the operators and users. The                    of the identified issues and provides a mapping of those to
    implementation of autonomy on UAV requires the                     the identified UAV’s functionalities. Diagrams of the
    application of innovative and at times new ways of                 architecture are also included. Finally, section 6 provides
    implementing the UAV on board functionalities,                     the conclusions of this study.
    specifically the need to implement artificial intelligence
    methodologies within the functional architecture of                2. Automatic versus Autonomous
    UAVs.
        This paper suggests a simple and highly modular UAV                The major driver for the adoption of unmanned
    system architecture for the on board UAV mission system            platforms is to have systems that can continuously provide
    and identifies a relationship between the functionalities          information and situation awareness without the risk of
    the UAV should present and the possible artificial                 losing human lives. The need of gathering information is
    intelligence methodologies that could be applied in order          driven by the need to make decisions and to react to the
    to achieve them.                                                   situation that presents itself as soon and effectively as
        The aim is to present a possible solution to the               possible.
    problem of autonomy on board of the platform and                       Current UAV systems are mostly examples of
    promote a hybrid approach to the issue of autonomous               automatic systems, which depend on pre-programmed



978-0-7695-3265-3/08 $25.00 © 2008 IEEE                          111
DOI 10.1109/BLISS.2008.14
flight and require constant monitoring from the human                    The payload functions are represented by the mission
operator (HO). Their use up to now has been described as             functional system, i.e. those systems which are mission
the use of “goggles in the sky” by a human operator on the           specific, and they are defined in [Ref. 1] (chp1, pp1), as
ground.                                                              those functions that include the set of all functions that
    By automatic is meant a repetitive action that does not          directly relate to the mission of a given vehicle
require external influence or control, but which repeats                 The payload of each platform is mission specific and it
itself based on some set conditions. A well known                    will vary with the specific mission requirement and it will
automatic system is feedback control system, which                   need to be fitted on the air vehicle and its control defined
automatically adjusts the input in order to obtain the               and implemented. Examples of possible sensors that could
desired outputs                                                      be fitted on a UAV are:
    The major difference between automatic and                       • Electro-Optical including Optical and IR sensors for
autonomous systems is that autonomous systems can                         obstacle detection and target detection
change their behaviour in response to unanticipated                  • Ground Moving Target Indicator (GMTI) radar and
events, whereas automatic systems would produce the                       Synthetic Aperture Radar (SAR)
same outputs regardless of any changes experienced by                • GPS/INS navigation system.
the system or its surroundings. In robotic applications,                 The vehicle management system (VMS), on the other
automatic actions allow the machines to be operationally             hand, is defined in [1] as the collection of functions that
autonomous, but they do not allow them to have                       are required for the vehicle to understand, plan, control,
decisional autonomy.                                                 and monitor the air vehicle operations. They usually
    An autonomous action or event is defined as an action            represent the safety critical functionalities required for the
or event that is ‘independent in mind or judgment’, a self-          safe employment of the platform; hence they include all
directed, self-governing entity, not controlled by others or         the flight-critical and safety-related functions.
by outside forces.                                                       The mission system functional capabilities can be
    Autonomous UAVs are herein defined as air vehicle                classified according to the level of autonomy that each of
systems with embedded autonomous functionalities.                    them would require if the HO was not included in the
    In this research the focus is on identifying                     loop.
methodologies to allow the level of autonomy for UAVs                    Three major levels of autonomy have been identified:
to increase and the implementation of a dynamic,                     • Low Level - The low level of autonomy can be
responsive behaviour on the platform itself to reduce the                 considered the reactive side of the UAV, where,
operator workload and provide independence from pre-                      given an event, the UAV will automatically react
programmed flight.                                                        according to pre-defined limits and following given
                                                                          dynamic models. This level is characterised by the
3. UAV Mission System                                                     group of functionalities that would be required in
                                                                          order to fly the platform remotely. These include, but
    The main gaps in current UAV systems are                              are not limited, to the flight control system, the
represented by the reliance on datalink, the ‘dumbness’ of                actuator function, the engine or propulsion control,
the control and payload systems, which means that they                    and the aircraft flight mechanics and air data
are not reactive to the environment, and the latency of                   acquisition.
information to the ground control station. In order to               • Medium Level – The medium level of autonomy
understand where the identified technological gaps can                    captures the ‘reflective’ capability that should be
present a significant drawback within the UAV systems,                    embedded on board of the platform in order to enable
the general UAV functionalities are captured.                             it to reason about its internal and external state. That
    A UAV mission system is seen to provide the                           is the ability of the platform to detect internal faults
translation of mission objectives into quantifiable,                      or malfunctions, to self-regulate when subjected to
scientific descriptions that provide a measure to judge the               unexpected events while carrying out the mission.
performance of the platform, i.e. the system in which the                 The functionalities of the mission systems that have
mission objectives are transformed into system                            been identified to belong to this class are:
parameters. According to the AGARD report on                             o     The flight path command and performance
Integrated Vehicle Management Systems [Ref. 1], the                            envelope protection, such as the waypoint
mission management functions of modern aerospace                               following system and the guidance and
vehicle can be split into two different functional elements:                   navigation functions;
the payload functions and the vehicle management                         o     The health manager and fault tolerant control, in
function.                                                                      order to detect and react to possible system
                                                                               failures and malfunctions;




                                                               112
o The power management system, included                      systems.
         optimising the power consumption and maximising             By developing those on the platform, the air vehicle is
         the mission time.                                       envisaged to achieve the required level of autonomy for
• High Level – The high level represents the most                the drone. The key capabilities and the associated system
      sophisticated layer of autonomy that it is desirable to    where it can be built in pointed out are reported in Table
      provide on board the air vehicle. It provides the          1. A possible application of these fields is provided in the
      platform with decision making capabilities and with        implementation of the systems reported in Table 1.
      the ability of interoperate with other platforms and /or
      systems in order to gather and analyse information, as
      well as the ability to reason and act upon the                    7DEOH /LQN EHWZHHQ 8$9 UHTXLUHG FDSDELOLWLHV DQG
      conclusions it draws. The functional systems which                           IXQFWLRQDO VXEVVWHPV
      have been identified to require a higher level of
      autonomy are:                                                       UAV capabilities                 Functional subsystem
    o the fault detection and identification (ID), i.e. the       React to changes in                 Mission Planner
         ability for the platform to detect the malfunction,      environment and be capable of       Mission Goal Manager
         reason about it and take an appropriate action           re-planning                         Sensor Manager
         without any HO intervention;                             Navigation through complex          Flight Path
    o the situation awareness manager, responsible to             terrain, possibly at high speed
         maintain and update the state of the world (i.e. the     Dynamic allocation of on board      Power Manager
         representation of the environment in which the           resources, i.e. fuel, power,        Sensor Manager
         UAV is operating) and communicate detected               sensors.
         changes to the mission manager.                          Interoperability with other         Sensor and Integrated Signal
    A high level of autonomy is also required to make             systems and platforms               and Data Processing Manager
decisions and to provide interoperability with other              Constant maintenance of             Sensor Manager, Integrated
systems. Therefore, the mission goal manager and sensor           situational awareness               Signal and Data Processing
manager, as well as the integrated signal and data                                                    Manager
processing managers fall into this layer. The mission goal                                            Situational awareness Manager
manager is responsible for redefining the mission                 Perform autonomous                  Mission Planner
objectives and goals and assessing their validity while           manoeuvre at the limits of the      Flight path envelope protection
cooperating with other air platforms or systems. The              flight envelope to maximise
sensor manager needs to monitor, select, and allocate the         performance
sensor when the target is detected and decide both the            Provide on board data and           Integrated Signal and Data
required accuracy and precision of information for the            information processing              Processing Manager
given task, and also the data that need to be recorded in         capability
order to successfully support the mission execution. Data         Operate outside communication       Mission Goal Manager
may come from the own sensor platforms mounted on the             link limits                         Mission Planner
UAV or from other air vehicles or systems. The integrated         Sense and avoid threats,            Sensor Manager
signal and data processing manager has the task of                collision avoidance                 Mission Planner
processing the multiple source information, including the         Autonomous reconfiguration of       Fault Tolerant/Reconfigurable
operator commands and reasoning about their soundness             systems in case of fault            control
and applicability within the mission context, and also to                                             Fault Detection and ID Manager
process them into appropriate information useful by the
other subsystems. Figure 1 captures the envisaged mission        5. Artificial intelligence techniques applicable
system architecture and the three major autonomy levels
                                                                 to UAVs
as described above.
    It is important to observe that the application of
                                                                     Conventional control systems and deterministic
artificial intelligence technique is envisaged to be used
                                                                 optimisation techniques provide a means to deal with
mainly in the high-level autonomy layer.
                                                                 uncertainty within certain boundaries. However, the
                                                                 multitude of case scenarios faced by the UAV even before
4. Identified mission level functionalities for                  it starts its mission requires a real time adaptive system
UAVs                                                             capable of reacting to unforeseen events to the best of its
                                                                 possibilities. Machines are now required to ‘think’ and
  In this section the key mission-level capabilities for a       ‘learn’ from the environment, as it has been pointed out in
UAV are discussed and linked to the above functional             the previous paragraphs.




                                                                  113
Surveillance/IFF

                                                      UAV Mission System
                                                                                                     Target Detection and Tracking
                       Mission Objectives

                                                                                                     Battle Damage
                                                                                                     Assessment/Reconnaissance/
                                                                                                     Search and Rescue
                            Mission
                                                                                                     )LUH &RQWURO :HDSRQ
                         requirements
                                                                                                     6VWHP

                                                                                                     Mission Communication
                         Mission Design                       Payload Functions
                          Integration
                                                                                                     Sensor Control/Sensor Fusion



                                                             Vehicle Management                      Aerial Refuelling/Rearming
                                                                  Functions
                                                                                                     Deployment of Humanitarian
                                                                                                     Aids



              /RZ /HYHO $XWRQRP                      0HGLXP /HYHO $XWRQRP                                         $XWRQRP
                                                                                                         +LJK /HYHO $XWRQRP
         •   )OLJKW &RQWURO                 -   )OLJKW 3DWK &RPPDQGV (QYHORSH                 • )DXOW 'HWHFWLRQ DQG ,'
         •   )OLJKW PHFKDQLFV                   3URWHFWLRQ                                    • 0LVVLRQ 3ODQQHU
         •   $FWXDWRU FRQWURO               -   0XOWLIXQFWLRQ ,QWHJUDWHG 1DYLJDWLRQ           • 6LWXDWLRQ $ZDUHQHVV 0DQDJHU
         •   3URSXOVLRQ (QJLQH                  6VWHP                                        'HFLVLRQ 0DNLQJ VXEVVWHPV
             &RQWURO                        -   +HDOWK 0DQJHU                                 • 0LVVLRQ *RDO 0DQDJHU
                                            -   )DXOW 7ROHUDQW 5HFRQILJXUDEOH &RQWURO         • 6HQVRU 0DQDJHU
                                            -   3RZHU 0DQDJHPHQW                              • ,QWHJUDWHG 6LJQDO DQG 'DWD 3URFHVVLQJ
                                                                                                0DQDJHU



                                                                               •    Monitoring – identify changes in a system’s observed
    )LJXUH 8$9 RQ ERDUG PLVVLRQ VVWHPV                                             state;
Therefore, the introduction of artificial intelligence                         • Risk Analysis – identify issues within a given course
methodology in order to implement, enhance, and improve                             of action and plan/provide mitigations, such as
the UAV’s autonomous functionalities seems to be a                                  alternative routes in case of detected obstacles;
natural evolution of more conventional feedback control                        • Data Analysis/Processing – for instance data mining,
theory.                                                                             identification of trends, extrapolation of data;
   The following branches of AI seem to be the most                            • Optimisation – streamlining a system or object, such
significant for potentially providing autonomous                                    as the route plan or fuel consumption for given
capabilities for the air vehicle. They were adapted                                 manoeuvre, to achieve the best performance, resource
partially from [2], pp. 248, and they are as follows:                               allocation;
• Interpretation – consists of data analysis coupled with                      • Classification – Assignment of a category to an
     domain knowledge forming high-level conclusions;                               object, for instance threat identification;
• Prediction – projecting probable consequences of                             • Control of Systems – Governing the behaviour of
     given situations;                                                              complex systems. Manipulation of a system’s
• Diagnostic – Determining the cause of malfunctions                                interaction with the world adjusting (actuating) the
     in complex situations based on observable symptoms.                            control surfaces, for instance to maintain a flight path
     Identify abnormalities in the observed states of the                         The applicability of those fields to the functional
      UAV’s systems and possibly suggest remedies to                           subsystems previously outlined is shown in the following
      mitigate fault;                                                          table and it is based on the author’s engineering judgment.
• Design – finding a configuration of system
     components, which can meet the required                                       7DEOH $SSOLFDWLRQ RI $, ILHOGV· WR IXQFWLRQDO VXEVVWHPV
     performance while satisfying the constraints;                                    AI Field                       Functional Subsystem
• Planning/Scheduling – devise a series of actions to                          Interpretation               •       Mission Planner
     achieve certain goal and co-ordinate them                                                              •       Situation Awareness Manager
     sequentially, for instance sensor pointing time or                                                     •       Fault Detection and ID
     route planning;                                                                                        •       Mission Goal Manager
• Decision Support/Decision Making – advise the HO,                                                         •       Sensor Manager
     both on the ground or on a fighter aircraft acting as                     Prediction                   •       Mission Planner
     mission manager, to aid them in difficult cognitive
                                                                                                            •       Situation Awareness Manager
     tasks;
                                                                                                            •       Fault Detection and ID




                                                                         114
•   Mission Goal Manager                    no longer sufficient to provide the additional capabilities
                         •   Sensor Manager                          of the UAV, it is still crucial to the implementation of
Diagnostic               •   Fault Detection and ID                  adaptive and reconfigurable control subsystems. Also,
                         •   Fault                                   optimisation is more and more being adopted for
                             Tolerant/Reconfigurable                 subsystems requiring resource allocation, planning and
                             Control                                 scheduling. In fact, heuristic searches and genetic
                         •   Health Manager                          algorithms are found to be well suited providing a quick
                                                                     and optimal solution when faced with multiple variable
Planning and             •   Mission Planner
                                                                     scenarios and incomplete information.
Scheduling               •   Mission Goal Manager
                                                                         Over the years, several AI techniques have been
                         •   Sensor Manager
                                                                     developed, each claiming to provide significant
Decision                 •   Mission Planner                         advantages over the others. No individual technique has
Support/Making           •   Situation Awareness Manager             proved to be the answer to the problem of creating
                         •   Fault Detection and ID                  machine autonomy. Therefore, it is necessary to blend the
                         •   Mission Goal Manager                    different methodologies and provide a new level of
                         •   Sensor Manager                          integration in order to create hybrid systems. Only by
Monitoring               •   Mission Planner                         combining different methodologies and matching them to
                         •   Situation Awareness Manager             the systems requirements it will be possible to move
                         •   Fault Detection and ID                  autonomy forward.
                         •   Mission Goal Manager                        In this research, the focus has been on identifying
                         •   Sensor Manager                          methodologies that could allow an increased capability for
                         •   Fault                                   a dynamic, responsive behaviour and its implementation
                             Tolerant/Reconfigurable                 on the platform itself so to reduce the operator workload
                             Control                                 and provide independence from pre-programmed flight.
                         •   Health Manager                          Therefore, this paper has focused on techniques, which
Risk Analysis            •   Mission Planner                         are believed to be matured enough or most appropriate for
                         •   Situation Awareness Manager             the application to UAV.
                                                                         The AI methodologies selected to provide autonomy to
                         •   Fault Detection and ID
                                                                     the UAV were:
                         •   Mission Goal Manager
                                                                          •   Artificial Neural Networks (ANN or NN) ([7],
                         •   Sensor Manager
                                                                              [2],[11]);
Data Analysis &          •   Sensor Manager                               •   Fuzzy Logic (Fuzzy) ([16],);
Processing               •   Integrated Signal and Data                   •   Genetic Algorithms (GA) (16], [11], [2]);
                             Processing Manager                           •   Reinforcement Learning (RL) ([4], [8], [13]);
Optimisation             •   Sensor Manager                               •   Temporal Logic (TL) ([15], [2]);
                         •   Power Manager                                •   Knowledge Based Systems (KBS);
                         •   Situation Awareness Manager                  •   Rule Base Systems (RBS) [14], [10], [11]);
Classification           •   Mission Planner                              •   Case Based Reasoning (CBR) ([9], [11], [2]);
                         •   Situation Awareness Manager                  •   Constrain Satisfaction Problem (CSP) ([3], [11]);
                         •   Fault Detection and ID                       •   Model Based Reasoning (MBR) ([2], [11]).
                         •   Mission Goal Manager                         In order to understand what techniques could be
                         •   Sensor Manager                          applied to the given AI fields in order to enable more
Control of Systems       •   Health Manager                          autonomous functionalities on board of the platform, the
                         •   Flight Protection                       selected techniques were tabulated against the identified
                             Commands/Envelope                       AI fields and they were rated according to their
                             protection                              applicability. The following table (Table 3) shows how
                         •   Multifunction Integrated                each technique could contribute to the set tasks which
                             Navigation System                       need to be performed.
                         •   Fault tolerant/Reconfigurable                In order to provide interpretation and diagnostic
                             Control                                 capabilities to the UAV systems, it is important to
                                                                     introduce data fusion techniques and data mining
                                                                     procedures to quickly process and analyse the data.
   It is important to observe that each subsystem requires
                                                                     Information and data fusion become particularly important
multiple AI fields in order to satisfy the required level of
                                                                     in the sensor management system, where the sensor
autonomy and that even though control theory in itself is
                                                                     information are collected and processed in order to




                                                               115
provide the current world state situation awareness and       advantages and drawbacks of implementing the individual
          update the belief functions on board of the UAV.              systems with the suggested techniques and analyse the
              The following is a qualitative and judgmental analysis.   issues associated with their integration. Moreover, by
             The analysis carried out in the previous paragraphs led    implementing specific systems, different techniques may
          to the definition of a functional UAV mission system          become more appealing compared to the one suggested.
          architecture with possible AI techniques associated with         Therefore, more research should be carried out on
          those subsystems. The result is shown in Figure 2 and         overall UAV functional system design and to their
          Figure 3. Figure 2 captures the high level decision making    integration. The key difference to current systems is the
          functionalities, usually performed by the pilot or the HO.    inclusion of artificial intelligence techniques within the
          Figure two represents the subsystems such as the guidance     functional domain.
          system that require a medium level of autonomy and the
          most inner part of the control loop, which can be fully       7. References
          automatic. The architecture has been subdivided into three
          components each of which groups together the functional       [1] AGARD Advisory Report, ‘Integrated Vehicle Management
          subsystems according to their required autonomy level             System’, AR 343, NATO, April 1996.
          and to their functional flow of information.                  [2] Luger, G. F., ‘Artificial Intelligence: Structures and
             The AI techniques selected for each subsystem are              Strategies for Complex Problem Solving’, Addison Wesley,
          drawn from Table 3.                                               IV Edition, 2002
                                                                        [3] Barták, R., ‘Constraint Processing’, IJCAI_07 Tutorial
                                                                        [4] Berenji, H. R., et alt.,‘Co-evolutionary Perception-based
                  7DEOH 0DSSLQJ RI $, 7HFKQLTXHV WR $, )LHOGV               Reinforcement Learning For Sensor Allocation in
                                                                            Autonomous Vehicles’, IEEE, 2003
                 NN Fuzzy GA RL TL KBS RBS CBR CSP MBS                  [5] Dufrene, W. R., ‘Approach for Autonomous Control of
Scheduling/                                                                 Unmanned Aerial Vehicle Using Intelligent Agents for
Planning                                                                    Knowledge Creation’, IEEE, 2004
                                                                        [6] Grelle, C., Ippolito, L., Loia, V., and Siano, P., ‘Agent –
Decision
                                                                            based architecture for designing hybrid control systems’,
Support/
                                                                            Information Sciences, Vol. 176, pp. 1103-1130, 2006.
Decision
                                                                        [7] Haykin, S., ‘Neural Networks: A comprehensive
Making
                                                                            Foundation’, Prentice Hall, II Edition
Diagnostic                                                              [8] Harmon, M. E., and Harmon, S., ‘Reinforcement Learning:
Risk analysis                                                               A Tutorial’.
                                                                        [9] Lees, B., ‘9th UK Workshop on Case-Based Reasoning:
Data                                                                        Proceeding’, SGAI, December 2004
Analysis/                                                               [10] Nilsson, N. J., ‘Introduction to Machine Learning: An early
Processing                                                                  draft of a proposed textbook’, Robotics Laboratory, Dep. Of
Monitor                                                                     Computer Science, Stanford University, Dec 1996
Optimisation                                                            [11] Russell, S. and Norvig, P., ‘Artificial Intelligence: A
                                                                            Modern Approach’, Prentice Hall, 2003
Interpretation                                                          [12] Shirazi, M. A., and Soroor, J., ‘An intelligent agent-based
Classification                                                              architecture for strategic information system application’.
                                                                            Knowledge Based Systems, 2006.
Control                                                                 [13] Ten Hagen, S. and Kröse, B., ‘A Short Introduction to
System,                                                                     Reinforcement Learning’, 7th Belgian-Dutch Conference on
Prediction                                                                  Machine Learning, pp 7-12, 1997
                                                                        [14] Tunstel, E., et al., ‘Rule-based reasoning and neural
Design
                                                                            networks for safe off-road robot mobility’, Expert Systems,
                                                                            Vol.19, No 4, September 2002
          Legend:                                                       [15] Vila, L., ‘A Survey on Temporal Reasoning in Artificial
                                                                            Intelligence’, AI Communications, Vol. 7, No. 1, March
                  Highly applicable
                                                                            1994
                  Potentially applicable                                [16] Johnson, J. and Picton, P., ‘Concepts in Artificial
                  Not applicable                                            Intelligence: designing Intelligent Machines, Volume 2’,
                                                                            Butterworth-Heinemann editions in association with the
                                                                            Open University, 2001
          7. Future work
             There are several issues that have not been addressed
          in this paper. Future research should investigate the




                                                                116
Mission                    Mission
                         Objectives                Requirements
                                                                                                           Fault Detection
                                                                                                                & ID

                               Mission Goal Manager                                                          CBR & CSP

                            GOAL Manager:
                                •     Update Goals;                                                                              Flight Path Commands
                                •     Eliminate     the     one
                                      reached/unachievable;                                                Mission Planner
                                •     Generate new goals
                                    CBR, CSP, RL, GA                                                       CSP, RL, GA



               PAYLOAD                           Integrated Signal and Data Processing

                                                    Integrated
                    INS/                              Signal                Integrated                         Situation
                    GPS                             Processing              Data Fusion                       Awareness
                                                                            Processing                         Manager
                                                  OUTPUTS:
                    Radio                         • Range                                                    KBS, GA, RL
                    Alt




117
                                                  • Range Rate                KBS
                                                  • Angle                   ANN, GA,
                                                  • Potential ID,
                    EO                              Size, Shape
                                                                              RL
                                                                                             World KBS         MBS on Air
                                                                                              Database          vehicle
                    SAR/
                    GMTI


                                                                                                         High Level Control On-Board PC



                                                        Sensor Manger
                                                                                              Fault
                                                          TL, RL, CBR, GA                  Detection &
                                                                                               ID
                                                                                          CBR, GA, NN                MBS on Sensor



      )LJXUH $, WHFKQLTXHV DVVRFLDWHG WR 8$9 IXQFWLRQDO VXEVVWHPV UHTXLULQJ KLJK OHYHO RI DXWRQRP
Mid-Level Command                   Terrain database

       Commands from
       Mission Manager                                                            Multifunction
                              Flight Path Commands                            integrated navigation   UAV States
                                                                                 Adaptive Mode
                               Trajectory Generator                                Transition
                              RL, CSP, TL                                    KBS, RL, TL, ANN



                                   Flight Control/                                UAV States
                               Loads/Propulsion/Flight
                                     Mechanics                                  Flight
                                                                                                                                           Low Level Flight
                                                                           mechanics/Propulsi                                                 Control
                               Envelope Protection                         on/FCS/ Actuators
                                    RL, CSP                                                                                                      Actuato
                                                                                                                                        FCS      r Model       UAV
                                                                                                                                        NN,                    6-
                                                                                                                                        GA,                    DOF
                                                                                                                                        Fuzzy




118
                                                                                                                                                               Model
                                       Flight                               Health Manager
                               Mechanics/Propulsion/                         CBS, MBS, TL                                                             Sensor
                               FCS/Electrical/Air Data                                                                                                Inputs
                                                                                                                                        Engine
                                                                                                                                        Model
                                                                                                                   DATALINK/INTERFACE




                               Fault Tolerant Control
                                  Reconfiguration
                              MBS & CBS, TL, ANN,
                              Fuzzy
                                                                     UAV States



                   MBS for          MBS for              MBS for
                    FCS             Engine               Actuators




      )LJXUH $, WHFKQLTXHV DVVRFLDWHG WR 8$9 IXQFWLRQDO VXEVVWHPV UHTXLULQJ PHGLXP DQG ORZ OHYHO DXWRQRP

More Related Content

Similar to Ai in decision making capability on board unmanned aerial vehicle

Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...Editor IJCATR
 
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...Editor IJCATR
 
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...IOSR Journals
 
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...IOSR Journals
 
A review on distributed control of
A review on distributed control ofA review on distributed control of
A review on distributed control ofijaia
 
Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...
Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...
Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...Angelo State University
 
IRJET- New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...
IRJET- 	  New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...IRJET- 	  New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...
IRJET- New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...IRJET Journal
 
System Architecture Study Global Hawk Unamanned Aerial System (UAS)
System Architecture Study Global Hawk Unamanned Aerial System (UAS)System Architecture Study Global Hawk Unamanned Aerial System (UAS)
System Architecture Study Global Hawk Unamanned Aerial System (UAS)University of Southern California
 
Vizer_MSc_Thesis_2011
Vizer_MSc_Thesis_2011Vizer_MSc_Thesis_2011
Vizer_MSc_Thesis_2011Daniel Vizer
 
Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...
Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...
Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...IJECEIAES
 
Drone simulators, advancements and challenges
Drone simulators, advancements and challengesDrone simulators, advancements and challenges
Drone simulators, advancements and challengesNile University
 
IFAC97.doc
IFAC97.docIFAC97.doc
IFAC97.docbutest
 
IFAC97.doc
IFAC97.docIFAC97.doc
IFAC97.docbutest
 
verification of autonomous robotic system
verification of autonomous robotic systemverification of autonomous robotic system
verification of autonomous robotic systemASJAYASURYA
 

Similar to Ai in decision making capability on board unmanned aerial vehicle (20)

Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
 
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
Software Architecture Evaluation of Unmanned Aerial Vehicles Fuzzy Based Cont...
 
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
 
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
Design and Structural Analysis for an Autonomous UAV System Consisting of Sla...
 
A review on distributed control of
A review on distributed control ofA review on distributed control of
A review on distributed control of
 
Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...
Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...
Detection of Sensor Faults in Small Helicopter UAVs Using Observer/Kalman Fil...
 
IRJET- New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...
IRJET- 	  New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...IRJET- 	  New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...
IRJET- New Approach to a Hybrid Fuzzy-Sliding Mode Control to a Brushless ...
 
System Architecture Study Global Hawk Unamanned Aerial System (UAS)
System Architecture Study Global Hawk Unamanned Aerial System (UAS)System Architecture Study Global Hawk Unamanned Aerial System (UAS)
System Architecture Study Global Hawk Unamanned Aerial System (UAS)
 
Apres Cobem09
Apres Cobem09Apres Cobem09
Apres Cobem09
 
Vizer_MSc_Thesis_2011
Vizer_MSc_Thesis_2011Vizer_MSc_Thesis_2011
Vizer_MSc_Thesis_2011
 
Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...
Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...
Automation of Air Traffic Management Using Fuzzy Logic Algorithm to Integrate...
 
SADP-UNIT-VI.pdf
SADP-UNIT-VI.pdfSADP-UNIT-VI.pdf
SADP-UNIT-VI.pdf
 
Drone simulators, advancements and challenges
Drone simulators, advancements and challengesDrone simulators, advancements and challenges
Drone simulators, advancements and challenges
 
Formalatc 287
Formalatc 287Formalatc 287
Formalatc 287
 
RADAR, Mlat, ADS, Bird RADAR, Weather RADAR Guide
RADAR, Mlat, ADS, Bird RADAR, Weather RADAR GuideRADAR, Mlat, ADS, Bird RADAR, Weather RADAR Guide
RADAR, Mlat, ADS, Bird RADAR, Weather RADAR Guide
 
avionics-architectures1.ppt
avionics-architectures1.pptavionics-architectures1.ppt
avionics-architectures1.ppt
 
IFAC97.doc
IFAC97.docIFAC97.doc
IFAC97.doc
 
IFAC97.doc
IFAC97.docIFAC97.doc
IFAC97.doc
 
verification of autonomous robotic system
verification of autonomous robotic systemverification of autonomous robotic system
verification of autonomous robotic system
 
Helicopter With Gps
Helicopter With GpsHelicopter With Gps
Helicopter With Gps
 

More from Dinesh More

Fuzzy logic and the goals of artificial intelligence
Fuzzy logic and the goals of artificial intelligenceFuzzy logic and the goals of artificial intelligence
Fuzzy logic and the goals of artificial intelligenceDinesh More
 
Event driven, mobile artificial intelligence algorithms
Event driven, mobile artificial intelligence algorithmsEvent driven, mobile artificial intelligence algorithms
Event driven, mobile artificial intelligence algorithmsDinesh More
 
Artificial intelligence(simulating the human mind)
Artificial intelligence(simulating the human mind)Artificial intelligence(simulating the human mind)
Artificial intelligence(simulating the human mind)Dinesh More
 
Artificial intelligence in mobile learning
Artificial intelligence in mobile learningArtificial intelligence in mobile learning
Artificial intelligence in mobile learningDinesh More
 
Artificial intelligence in cyber defense
Artificial intelligence in cyber defenseArtificial intelligence in cyber defense
Artificial intelligence in cyber defenseDinesh More
 
(Ai) in power transformer fault diagnosis
(Ai) in power transformer fault diagnosis(Ai) in power transformer fault diagnosis
(Ai) in power transformer fault diagnosisDinesh More
 
Artificial Intelligence
Artificial IntelligenceArtificial Intelligence
Artificial IntelligenceDinesh More
 

More from Dinesh More (7)

Fuzzy logic and the goals of artificial intelligence
Fuzzy logic and the goals of artificial intelligenceFuzzy logic and the goals of artificial intelligence
Fuzzy logic and the goals of artificial intelligence
 
Event driven, mobile artificial intelligence algorithms
Event driven, mobile artificial intelligence algorithmsEvent driven, mobile artificial intelligence algorithms
Event driven, mobile artificial intelligence algorithms
 
Artificial intelligence(simulating the human mind)
Artificial intelligence(simulating the human mind)Artificial intelligence(simulating the human mind)
Artificial intelligence(simulating the human mind)
 
Artificial intelligence in mobile learning
Artificial intelligence in mobile learningArtificial intelligence in mobile learning
Artificial intelligence in mobile learning
 
Artificial intelligence in cyber defense
Artificial intelligence in cyber defenseArtificial intelligence in cyber defense
Artificial intelligence in cyber defense
 
(Ai) in power transformer fault diagnosis
(Ai) in power transformer fault diagnosis(Ai) in power transformer fault diagnosis
(Ai) in power transformer fault diagnosis
 
Artificial Intelligence
Artificial IntelligenceArtificial Intelligence
Artificial Intelligence
 

Ai in decision making capability on board unmanned aerial vehicle

  • 1. Bio-inspired, Learning and Intelligent systems for security Bio-inspired, learning intelligent Systems for Security Artificial intelligence methodologies applicable to support the decision-making capability on board Unmanned Aerial Vehicles Isabella Panella Thales UK, Aerospace Division, Manor Royal, Crawley, UK isabella.panella@uk.thalesgroup.com Abstract systems, as single methodologies cannot address The need for Unmanned Air Vehicles (UAVs) to operate independently the complexity of the problem of machine autonomously and to manage their operation with autonomy. minimal intervention from the ground control station, in The architecture specifies the relationship between the order to reduce the datalink utilization and maximize various subsystems that are seen to be key enablers to the their exploitation in beyond line of sight (BLOS) autonomy of the vehicle combined with legacy systems operations, has been long recognized within industry and and the artificial intelligence (AI) techniques that could be research institutes. Many artificial intelligence (AI) used in order to implement them. techniques try to address the challenge of moving UAV The motivation behind this work is to provide a towards full autonomy. However, no single technique has possible UAV functional architecture within which the AI been able to provide the required autonomy for methodologies can be embedded and the need for them to unmanned platforms. This paper presents a Unmanned complement each others weaknesses and interrelationship Air Systems (UAS) architecture within which the different appreciated. AI methodologies applicable to each subsystem are In this paper, unmanned air platforms refer to both the presented. air vehicle and the payloads fitted on the air vehicle in order to provide the required capabilities for the given mission. 1. Introduction The paper is organised as follows. Section 2 provides a definition of automatic versus autonomous. These terms It has long being recognized that the employment of are often confused and misused and it is appropriate to be Unmanned Air Systems (UAS) could provide significant clear about the difference between them. Section 3 advantages in military applications for ‘dirty, dull, and provides an outline of the major functional elements of the dangerous’ missions as well as for civil and commercial platform and the payload is also provided. applications, such as search and rescue, border Section 4 reports some of the key mission level management, pipeline monitoring. functionalities and capabilities that are required for the UAV performance can be improved by increasing the systems to work autonomously. Section 5 identifies the AI autonomy on-board of the platforms in order to reduce the fields and techniques that can be applicable to solve some workload placed on the operators and users. The of the identified issues and provides a mapping of those to implementation of autonomy on UAV requires the the identified UAV’s functionalities. Diagrams of the application of innovative and at times new ways of architecture are also included. Finally, section 6 provides implementing the UAV on board functionalities, the conclusions of this study. specifically the need to implement artificial intelligence methodologies within the functional architecture of 2. Automatic versus Autonomous UAVs. This paper suggests a simple and highly modular UAV The major driver for the adoption of unmanned system architecture for the on board UAV mission system platforms is to have systems that can continuously provide and identifies a relationship between the functionalities information and situation awareness without the risk of the UAV should present and the possible artificial losing human lives. The need of gathering information is intelligence methodologies that could be applied in order driven by the need to make decisions and to react to the to achieve them. situation that presents itself as soon and effectively as The aim is to present a possible solution to the possible. problem of autonomy on board of the platform and Current UAV systems are mostly examples of promote a hybrid approach to the issue of autonomous automatic systems, which depend on pre-programmed 978-0-7695-3265-3/08 $25.00 © 2008 IEEE 111 DOI 10.1109/BLISS.2008.14
  • 2. flight and require constant monitoring from the human The payload functions are represented by the mission operator (HO). Their use up to now has been described as functional system, i.e. those systems which are mission the use of “goggles in the sky” by a human operator on the specific, and they are defined in [Ref. 1] (chp1, pp1), as ground. those functions that include the set of all functions that By automatic is meant a repetitive action that does not directly relate to the mission of a given vehicle require external influence or control, but which repeats The payload of each platform is mission specific and it itself based on some set conditions. A well known will vary with the specific mission requirement and it will automatic system is feedback control system, which need to be fitted on the air vehicle and its control defined automatically adjusts the input in order to obtain the and implemented. Examples of possible sensors that could desired outputs be fitted on a UAV are: The major difference between automatic and • Electro-Optical including Optical and IR sensors for autonomous systems is that autonomous systems can obstacle detection and target detection change their behaviour in response to unanticipated • Ground Moving Target Indicator (GMTI) radar and events, whereas automatic systems would produce the Synthetic Aperture Radar (SAR) same outputs regardless of any changes experienced by • GPS/INS navigation system. the system or its surroundings. In robotic applications, The vehicle management system (VMS), on the other automatic actions allow the machines to be operationally hand, is defined in [1] as the collection of functions that autonomous, but they do not allow them to have are required for the vehicle to understand, plan, control, decisional autonomy. and monitor the air vehicle operations. They usually An autonomous action or event is defined as an action represent the safety critical functionalities required for the or event that is ‘independent in mind or judgment’, a self- safe employment of the platform; hence they include all directed, self-governing entity, not controlled by others or the flight-critical and safety-related functions. by outside forces. The mission system functional capabilities can be Autonomous UAVs are herein defined as air vehicle classified according to the level of autonomy that each of systems with embedded autonomous functionalities. them would require if the HO was not included in the In this research the focus is on identifying loop. methodologies to allow the level of autonomy for UAVs Three major levels of autonomy have been identified: to increase and the implementation of a dynamic, • Low Level - The low level of autonomy can be responsive behaviour on the platform itself to reduce the considered the reactive side of the UAV, where, operator workload and provide independence from pre- given an event, the UAV will automatically react programmed flight. according to pre-defined limits and following given dynamic models. This level is characterised by the 3. UAV Mission System group of functionalities that would be required in order to fly the platform remotely. These include, but The main gaps in current UAV systems are are not limited, to the flight control system, the represented by the reliance on datalink, the ‘dumbness’ of actuator function, the engine or propulsion control, the control and payload systems, which means that they and the aircraft flight mechanics and air data are not reactive to the environment, and the latency of acquisition. information to the ground control station. In order to • Medium Level – The medium level of autonomy understand where the identified technological gaps can captures the ‘reflective’ capability that should be present a significant drawback within the UAV systems, embedded on board of the platform in order to enable the general UAV functionalities are captured. it to reason about its internal and external state. That A UAV mission system is seen to provide the is the ability of the platform to detect internal faults translation of mission objectives into quantifiable, or malfunctions, to self-regulate when subjected to scientific descriptions that provide a measure to judge the unexpected events while carrying out the mission. performance of the platform, i.e. the system in which the The functionalities of the mission systems that have mission objectives are transformed into system been identified to belong to this class are: parameters. According to the AGARD report on o The flight path command and performance Integrated Vehicle Management Systems [Ref. 1], the envelope protection, such as the waypoint mission management functions of modern aerospace following system and the guidance and vehicle can be split into two different functional elements: navigation functions; the payload functions and the vehicle management o The health manager and fault tolerant control, in function. order to detect and react to possible system failures and malfunctions; 112
  • 3. o The power management system, included systems. optimising the power consumption and maximising By developing those on the platform, the air vehicle is the mission time. envisaged to achieve the required level of autonomy for • High Level – The high level represents the most the drone. The key capabilities and the associated system sophisticated layer of autonomy that it is desirable to where it can be built in pointed out are reported in Table provide on board the air vehicle. It provides the 1. A possible application of these fields is provided in the platform with decision making capabilities and with implementation of the systems reported in Table 1. the ability of interoperate with other platforms and /or systems in order to gather and analyse information, as well as the ability to reason and act upon the 7DEOH /LQN EHWZHHQ 8$9 UHTXLUHG FDSDELOLWLHV DQG conclusions it draws. The functional systems which IXQFWLRQDO VXEVVWHPV have been identified to require a higher level of autonomy are: UAV capabilities Functional subsystem o the fault detection and identification (ID), i.e. the React to changes in Mission Planner ability for the platform to detect the malfunction, environment and be capable of Mission Goal Manager reason about it and take an appropriate action re-planning Sensor Manager without any HO intervention; Navigation through complex Flight Path o the situation awareness manager, responsible to terrain, possibly at high speed maintain and update the state of the world (i.e. the Dynamic allocation of on board Power Manager representation of the environment in which the resources, i.e. fuel, power, Sensor Manager UAV is operating) and communicate detected sensors. changes to the mission manager. Interoperability with other Sensor and Integrated Signal A high level of autonomy is also required to make systems and platforms and Data Processing Manager decisions and to provide interoperability with other Constant maintenance of Sensor Manager, Integrated systems. Therefore, the mission goal manager and sensor situational awareness Signal and Data Processing manager, as well as the integrated signal and data Manager processing managers fall into this layer. The mission goal Situational awareness Manager manager is responsible for redefining the mission Perform autonomous Mission Planner objectives and goals and assessing their validity while manoeuvre at the limits of the Flight path envelope protection cooperating with other air platforms or systems. The flight envelope to maximise sensor manager needs to monitor, select, and allocate the performance sensor when the target is detected and decide both the Provide on board data and Integrated Signal and Data required accuracy and precision of information for the information processing Processing Manager given task, and also the data that need to be recorded in capability order to successfully support the mission execution. Data Operate outside communication Mission Goal Manager may come from the own sensor platforms mounted on the link limits Mission Planner UAV or from other air vehicles or systems. The integrated Sense and avoid threats, Sensor Manager signal and data processing manager has the task of collision avoidance Mission Planner processing the multiple source information, including the Autonomous reconfiguration of Fault Tolerant/Reconfigurable operator commands and reasoning about their soundness systems in case of fault control and applicability within the mission context, and also to Fault Detection and ID Manager process them into appropriate information useful by the other subsystems. Figure 1 captures the envisaged mission 5. Artificial intelligence techniques applicable system architecture and the three major autonomy levels to UAVs as described above. It is important to observe that the application of Conventional control systems and deterministic artificial intelligence technique is envisaged to be used optimisation techniques provide a means to deal with mainly in the high-level autonomy layer. uncertainty within certain boundaries. However, the multitude of case scenarios faced by the UAV even before 4. Identified mission level functionalities for it starts its mission requires a real time adaptive system UAVs capable of reacting to unforeseen events to the best of its possibilities. Machines are now required to ‘think’ and In this section the key mission-level capabilities for a ‘learn’ from the environment, as it has been pointed out in UAV are discussed and linked to the above functional the previous paragraphs. 113
  • 4. Surveillance/IFF UAV Mission System Target Detection and Tracking Mission Objectives Battle Damage Assessment/Reconnaissance/ Search and Rescue Mission )LUH &RQWURO :HDSRQ requirements 6VWHP Mission Communication Mission Design Payload Functions Integration Sensor Control/Sensor Fusion Vehicle Management Aerial Refuelling/Rearming Functions Deployment of Humanitarian Aids /RZ /HYHO $XWRQRP 0HGLXP /HYHO $XWRQRP $XWRQRP +LJK /HYHO $XWRQRP • )OLJKW &RQWURO - )OLJKW 3DWK &RPPDQGV (QYHORSH • )DXOW 'HWHFWLRQ DQG ,' • )OLJKW PHFKDQLFV 3URWHFWLRQ • 0LVVLRQ 3ODQQHU • $FWXDWRU FRQWURO - 0XOWLIXQFWLRQ ,QWHJUDWHG 1DYLJDWLRQ • 6LWXDWLRQ $ZDUHQHVV 0DQDJHU • 3URSXOVLRQ (QJLQH 6VWHP 'HFLVLRQ 0DNLQJ VXEVVWHPV &RQWURO - +HDOWK 0DQJHU • 0LVVLRQ *RDO 0DQDJHU - )DXOW 7ROHUDQW 5HFRQILJXUDEOH &RQWURO • 6HQVRU 0DQDJHU - 3RZHU 0DQDJHPHQW • ,QWHJUDWHG 6LJQDO DQG 'DWD 3URFHVVLQJ 0DQDJHU • Monitoring – identify changes in a system’s observed )LJXUH 8$9 RQ ERDUG PLVVLRQ VVWHPV state; Therefore, the introduction of artificial intelligence • Risk Analysis – identify issues within a given course methodology in order to implement, enhance, and improve of action and plan/provide mitigations, such as the UAV’s autonomous functionalities seems to be a alternative routes in case of detected obstacles; natural evolution of more conventional feedback control • Data Analysis/Processing – for instance data mining, theory. identification of trends, extrapolation of data; The following branches of AI seem to be the most • Optimisation – streamlining a system or object, such significant for potentially providing autonomous as the route plan or fuel consumption for given capabilities for the air vehicle. They were adapted manoeuvre, to achieve the best performance, resource partially from [2], pp. 248, and they are as follows: allocation; • Interpretation – consists of data analysis coupled with • Classification – Assignment of a category to an domain knowledge forming high-level conclusions; object, for instance threat identification; • Prediction – projecting probable consequences of • Control of Systems – Governing the behaviour of given situations; complex systems. Manipulation of a system’s • Diagnostic – Determining the cause of malfunctions interaction with the world adjusting (actuating) the in complex situations based on observable symptoms. control surfaces, for instance to maintain a flight path Identify abnormalities in the observed states of the The applicability of those fields to the functional UAV’s systems and possibly suggest remedies to subsystems previously outlined is shown in the following mitigate fault; table and it is based on the author’s engineering judgment. • Design – finding a configuration of system components, which can meet the required 7DEOH $SSOLFDWLRQ RI $, ILHOGV· WR IXQFWLRQDO VXEVVWHPV performance while satisfying the constraints; AI Field Functional Subsystem • Planning/Scheduling – devise a series of actions to Interpretation • Mission Planner achieve certain goal and co-ordinate them • Situation Awareness Manager sequentially, for instance sensor pointing time or • Fault Detection and ID route planning; • Mission Goal Manager • Decision Support/Decision Making – advise the HO, • Sensor Manager both on the ground or on a fighter aircraft acting as Prediction • Mission Planner mission manager, to aid them in difficult cognitive • Situation Awareness Manager tasks; • Fault Detection and ID 114
  • 5. Mission Goal Manager no longer sufficient to provide the additional capabilities • Sensor Manager of the UAV, it is still crucial to the implementation of Diagnostic • Fault Detection and ID adaptive and reconfigurable control subsystems. Also, • Fault optimisation is more and more being adopted for Tolerant/Reconfigurable subsystems requiring resource allocation, planning and Control scheduling. In fact, heuristic searches and genetic • Health Manager algorithms are found to be well suited providing a quick and optimal solution when faced with multiple variable Planning and • Mission Planner scenarios and incomplete information. Scheduling • Mission Goal Manager Over the years, several AI techniques have been • Sensor Manager developed, each claiming to provide significant Decision • Mission Planner advantages over the others. No individual technique has Support/Making • Situation Awareness Manager proved to be the answer to the problem of creating • Fault Detection and ID machine autonomy. Therefore, it is necessary to blend the • Mission Goal Manager different methodologies and provide a new level of • Sensor Manager integration in order to create hybrid systems. Only by Monitoring • Mission Planner combining different methodologies and matching them to • Situation Awareness Manager the systems requirements it will be possible to move • Fault Detection and ID autonomy forward. • Mission Goal Manager In this research, the focus has been on identifying • Sensor Manager methodologies that could allow an increased capability for • Fault a dynamic, responsive behaviour and its implementation Tolerant/Reconfigurable on the platform itself so to reduce the operator workload Control and provide independence from pre-programmed flight. • Health Manager Therefore, this paper has focused on techniques, which Risk Analysis • Mission Planner are believed to be matured enough or most appropriate for • Situation Awareness Manager the application to UAV. The AI methodologies selected to provide autonomy to • Fault Detection and ID the UAV were: • Mission Goal Manager • Artificial Neural Networks (ANN or NN) ([7], • Sensor Manager [2],[11]); Data Analysis & • Sensor Manager • Fuzzy Logic (Fuzzy) ([16],); Processing • Integrated Signal and Data • Genetic Algorithms (GA) (16], [11], [2]); Processing Manager • Reinforcement Learning (RL) ([4], [8], [13]); Optimisation • Sensor Manager • Temporal Logic (TL) ([15], [2]); • Power Manager • Knowledge Based Systems (KBS); • Situation Awareness Manager • Rule Base Systems (RBS) [14], [10], [11]); Classification • Mission Planner • Case Based Reasoning (CBR) ([9], [11], [2]); • Situation Awareness Manager • Constrain Satisfaction Problem (CSP) ([3], [11]); • Fault Detection and ID • Model Based Reasoning (MBR) ([2], [11]). • Mission Goal Manager In order to understand what techniques could be • Sensor Manager applied to the given AI fields in order to enable more Control of Systems • Health Manager autonomous functionalities on board of the platform, the • Flight Protection selected techniques were tabulated against the identified Commands/Envelope AI fields and they were rated according to their protection applicability. The following table (Table 3) shows how • Multifunction Integrated each technique could contribute to the set tasks which Navigation System need to be performed. • Fault tolerant/Reconfigurable In order to provide interpretation and diagnostic Control capabilities to the UAV systems, it is important to introduce data fusion techniques and data mining procedures to quickly process and analyse the data. It is important to observe that each subsystem requires Information and data fusion become particularly important multiple AI fields in order to satisfy the required level of in the sensor management system, where the sensor autonomy and that even though control theory in itself is information are collected and processed in order to 115
  • 6. provide the current world state situation awareness and advantages and drawbacks of implementing the individual update the belief functions on board of the UAV. systems with the suggested techniques and analyse the The following is a qualitative and judgmental analysis. issues associated with their integration. Moreover, by The analysis carried out in the previous paragraphs led implementing specific systems, different techniques may to the definition of a functional UAV mission system become more appealing compared to the one suggested. architecture with possible AI techniques associated with Therefore, more research should be carried out on those subsystems. The result is shown in Figure 2 and overall UAV functional system design and to their Figure 3. Figure 2 captures the high level decision making integration. The key difference to current systems is the functionalities, usually performed by the pilot or the HO. inclusion of artificial intelligence techniques within the Figure two represents the subsystems such as the guidance functional domain. system that require a medium level of autonomy and the most inner part of the control loop, which can be fully 7. References automatic. The architecture has been subdivided into three components each of which groups together the functional [1] AGARD Advisory Report, ‘Integrated Vehicle Management subsystems according to their required autonomy level System’, AR 343, NATO, April 1996. and to their functional flow of information. [2] Luger, G. F., ‘Artificial Intelligence: Structures and The AI techniques selected for each subsystem are Strategies for Complex Problem Solving’, Addison Wesley, drawn from Table 3. IV Edition, 2002 [3] Barták, R., ‘Constraint Processing’, IJCAI_07 Tutorial [4] Berenji, H. R., et alt.,‘Co-evolutionary Perception-based 7DEOH 0DSSLQJ RI $, 7HFKQLTXHV WR $, )LHOGV Reinforcement Learning For Sensor Allocation in Autonomous Vehicles’, IEEE, 2003 NN Fuzzy GA RL TL KBS RBS CBR CSP MBS [5] Dufrene, W. R., ‘Approach for Autonomous Control of Scheduling/ Unmanned Aerial Vehicle Using Intelligent Agents for Planning Knowledge Creation’, IEEE, 2004 [6] Grelle, C., Ippolito, L., Loia, V., and Siano, P., ‘Agent – Decision based architecture for designing hybrid control systems’, Support/ Information Sciences, Vol. 176, pp. 1103-1130, 2006. Decision [7] Haykin, S., ‘Neural Networks: A comprehensive Making Foundation’, Prentice Hall, II Edition Diagnostic [8] Harmon, M. E., and Harmon, S., ‘Reinforcement Learning: Risk analysis A Tutorial’. [9] Lees, B., ‘9th UK Workshop on Case-Based Reasoning: Data Proceeding’, SGAI, December 2004 Analysis/ [10] Nilsson, N. J., ‘Introduction to Machine Learning: An early Processing draft of a proposed textbook’, Robotics Laboratory, Dep. Of Monitor Computer Science, Stanford University, Dec 1996 Optimisation [11] Russell, S. and Norvig, P., ‘Artificial Intelligence: A Modern Approach’, Prentice Hall, 2003 Interpretation [12] Shirazi, M. A., and Soroor, J., ‘An intelligent agent-based Classification architecture for strategic information system application’. Knowledge Based Systems, 2006. Control [13] Ten Hagen, S. and Kröse, B., ‘A Short Introduction to System, Reinforcement Learning’, 7th Belgian-Dutch Conference on Prediction Machine Learning, pp 7-12, 1997 [14] Tunstel, E., et al., ‘Rule-based reasoning and neural Design networks for safe off-road robot mobility’, Expert Systems, Vol.19, No 4, September 2002 Legend: [15] Vila, L., ‘A Survey on Temporal Reasoning in Artificial Intelligence’, AI Communications, Vol. 7, No. 1, March Highly applicable 1994 Potentially applicable [16] Johnson, J. and Picton, P., ‘Concepts in Artificial Not applicable Intelligence: designing Intelligent Machines, Volume 2’, Butterworth-Heinemann editions in association with the Open University, 2001 7. Future work There are several issues that have not been addressed in this paper. Future research should investigate the 116
  • 7. Mission Mission Objectives Requirements Fault Detection & ID Mission Goal Manager CBR & CSP GOAL Manager: • Update Goals; Flight Path Commands • Eliminate the one reached/unachievable; Mission Planner • Generate new goals CBR, CSP, RL, GA CSP, RL, GA PAYLOAD Integrated Signal and Data Processing Integrated INS/ Signal Integrated Situation GPS Processing Data Fusion Awareness Processing Manager OUTPUTS: Radio • Range KBS, GA, RL Alt 117 • Range Rate KBS • Angle ANN, GA, • Potential ID, EO Size, Shape RL World KBS MBS on Air Database vehicle SAR/ GMTI High Level Control On-Board PC Sensor Manger Fault TL, RL, CBR, GA Detection & ID CBR, GA, NN MBS on Sensor )LJXUH $, WHFKQLTXHV DVVRFLDWHG WR 8$9 IXQFWLRQDO VXEVVWHPV UHTXLULQJ KLJK OHYHO RI DXWRQRP
  • 8. Mid-Level Command Terrain database Commands from Mission Manager Multifunction Flight Path Commands integrated navigation UAV States Adaptive Mode Trajectory Generator Transition RL, CSP, TL KBS, RL, TL, ANN Flight Control/ UAV States Loads/Propulsion/Flight Mechanics Flight Low Level Flight mechanics/Propulsi Control Envelope Protection on/FCS/ Actuators RL, CSP Actuato FCS r Model UAV NN, 6- GA, DOF Fuzzy 118 Model Flight Health Manager Mechanics/Propulsion/ CBS, MBS, TL Sensor FCS/Electrical/Air Data Inputs Engine Model DATALINK/INTERFACE Fault Tolerant Control Reconfiguration MBS & CBS, TL, ANN, Fuzzy UAV States MBS for MBS for MBS for FCS Engine Actuators )LJXUH $, WHFKQLTXHV DVVRFLDWHG WR 8$9 IXQFWLRQDO VXEVVWHPV UHTXLULQJ PHGLXP DQG ORZ OHYHO DXWRQRP