In this presentation, latest results of our research activity in the ESCEL Comp4Drones project is presented. The application of S3D to the modeling and performance analysis of drone-based services is described
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Villar presentation.pdf
1. June 8, 2022
Eugenio Villar
University of Cantabria
Modeling, Design and
Implementation of drone-based
Services
2. June 8, 2022
Agenda
▪ Introduction
▪ Model-Driven Design of CPSoS
▪ Drone-based Services
▪ Design Verification and Performance Analysis
▪ Experimental Results
▪ Conclusions
▪ Demos
▪ Slides can be found at:
▪ https://www.slideshare.net/EugenioVillar/
2
3. June 8, 2022
N2 B
N1
M A
Introduction
▪ Model-Driven Design (MDD)
▪ High-abstraction level
▪ Mature SW engineering methodology
▪ UML
▪ Standard, any (user-defined) MoC, any language
▪ Natural way to capture system architecture
3
▪ Semantic lacks
▪ Domain-specific profiles
▪ MetaMorph
▪ OpenSource, any (user-defined) MoC, language agnostic
4. June 8, 2022
Introduction
▪ Model-Driven Design (MDD)
▪ Abstraction of the platform
▪ SW development on APIs
▪ MWs, VMs, OSs….
▪ Simulation is not always an issue
▪ SiL & HiL Verification
▪ Performance is not key
4
M2C
PCs Servers DCs HPCs
Mobiles
IoT
compilation
MWs VMs OSs
Python Java C++ GO Swift…
5. June 8, 2022
Introduction
▪ State-of-the-Art in Simulation-Based MDD
▪ Matlab-Simulink
▪ Proprietary
▪ only one MoC, M language
▪ Application to UAVs
▪ Autopilot + Physics
▪ ROS toolbox
▪ AMeSIM/ANSyS
▪ Proprietary, only one MoC
▪ CoFluent
▪ Proprietary
▪ a few MoCs, C/C++ language
▪ Ptolemy II
▪ Academic, any MoC, C/C++ inside a Java block
▪ HEPSYCODE
▪ Academic, several MoCs, SystemC
▪ …
5
M2C
MWs VMs OSs
Python Java C++ GO Swift…
6. June 8, 2022
Introduction
▪ Embedded System Design (ESD) & Electronic System-Level Design (ESL)
▪ Model of the platform
▪ At different abstraction levels
▪ Platform-based HW/SW co-design
▪ Simulation is key
▪ At different abstraction levels
▪ Performance is key
▪ Cyber-Physical interaction
▪ Real-Time Systems
6
P&R
C++ SystemC VHDL/Verilog…
SoC
FPGAs
Boards
NoC
HW accel.
I/O Busses
GPUs
CPUs
MWs VMs OSs
HLS/RTL
synthesis
compilation
7. June 8, 2022
Introduction
▪ Services provided on computing platforms of many kind
▪ Programming the Internet of Everything
▪ In close interaction with the physical world => CPSoS-IoT
▪ Full abstraction of the computing platform is no longer possible
7
8. June 8, 2022
Introduction
▪ Gap between MDD & ESD/ESL
▪ S3D: Single-Source System Design
▪ Linking MDD with ESD/ESL
▪ Simulating the PSM
▪ At different abstraction levels
▪ Estimating performance
▪ Depending on the HW Platform
▪ Automatic generation of the executable models
▪ Design-Space Exploration
Python Java C++ GO Swift…
SoC
FPGAs
Boards
NoC
HW accel.
I/O Busses
GPUs
CPUs
MWs VMs OSs
??
Design
Gap
8
11. June 8, 2022
Introduction
▪ S3D: Design-Space Exploration
▪ Linking MDD with ESL
Python Java C++ GO Swift…
SoC
FPGAs
Boards
NoC
HW accel.
I/O Busses
GPUs
CPUs
MWs VMs OSs
11
12. June 8, 2022
Introduction
▪ S3D: Programming the Internet of Everything
▪ Services provided on heterogeneous computing platforms of many kind
▪ Impact on functional and non-functional properties of the execution platform
HW
Synthesis
SW
Synthesis
(RT)OS,
Runtime,…
Corba, TCP/IP,
MPI, MCAPI,…
ForTran,
Python, Java,
HTML5,
C/C++,…
Concerns
Performance,
safety, cost,
security, size,
…
Schedulability
Analysis
Simulation
Verification
Performance
Analysis
Design-Space
Exploration
Optimization
Service
UML
CPSoS DSL
UML
CPSoS DSL
12
13. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ Traditional V-Cycle for Mechatronic Systems
▪ Design & Analyze
▪ Develop (Platform-Dependent)
▪ Verify
13
14. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ New Validation/Verification V-Cycle
▪ Reusability
▪ Simulation-based Analysis & Design
▪ Design-Space Exploration
▪ Performance Analysis
▪ Architectural mapping
14
15. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ System Design & Validation
▪ System Interface
▪ Domain & System Requirements
▪ Functional Specification
▪ Input/Output Rates and Delays
▪ Test-Bench Development
▪ Minimal Functionality
▪ System Validation
15
16. June 8, 2022
▪ CPS: Digital Behavior in a Physical World
▪ System Model (Specification)
▪ The implementation is as good as
similar to the model
▪ Environment Model
▪ The model is as good as similar to reality
▪ Close-loop behavior can be extremely difficult to model
Model-Driven Design of Cyber-Physical SoS
16
17. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ Mechanical & HW/SW Design & Validation
▪ Sub-System & Component Interface
▪ Sub-System & Component Requirements
▪ Input/Output Rates and Delays
▪ Minimal Sub-System & Component Functionality => System Validation
▪ Reusability
18. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ ROS
▪ De-facto standard for robot applications
▪ ROS Infrastructure
▪ ROS nodes
▪ Processes that perform a certain computation
▪ ROS Master (Core)
▪ Nodes registration
▪ Communication manager
18
Client-Server (RPC) Publish-Suscriber
19. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ Platform-Based Design
▪ POSIX API
▪ ROS API
19
20. June 8, 2022
VIPPE Scheduler
Host Execution Platform
Ubuntu 18.04
Intel i5-34700 x86-64
Process 1 Process 2 Process 3
Model-Driven Design of Cyber-Physical SoS
▪ Platform-Based Simulation
▪ POSIX API
▪ ROS API
20
21. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ Drone-Independent ROS Interface
▪ Extensible to robots with similar functionality
21
22. June 8, 2022
Model-Driven Design of Cyber-Physical SoS
▪ Drone-Independent ROS Interface
▪ Reusable ROS code
▪ Code length reduction
▪ Decrease in simulation speed
22
Reduction in program length
Drone Drone-specific code (LoCs) CoDIn (LoCs) Reduction (%)
Ardupilot 343
233
32.07
Px4 357 34.73
Reduction in simulation speed
Drone Drone-specific code (s/s) CoDIn (s/s) Reduction (%)
Ardupilot 1.9 1,85 2,6
23. June 8, 2022
▪ S3D components in a drone-based service
▪ C++ components
▪ ROScpp components
▪ Drone and robot models
Design Verification & Performance Analysis
C++
ROScpp
Drone
models
23
24. June 8, 2022
▪ Multi-Level Simulation & Performance Analysis
▪ C++ and ROScpp components
▪ Simulation of ROScpp components without the (slow) ROS infrastructure
▪ Functional ROS
▪ Direct links among publishers and subscribers
Design Verification & Performance Analysis
Abstraction Levels for C++ & ROS cpp components
Level Code Timing/Energy ROS infrastructure
MN Minimal No
Yes/No
MC Minimal Constant
FC Full code Constant
FD Full code Data-dependent
24
25. June 8, 2022
▪ Native Simulation: Flaxibility + accuracy
…
Overflow = 0;
s = 1L;
for (i = 0; i < L_subfr; i++) {
Carry = 0;
s = L_macNs(s, xn[i], y1[i]);
if (Overflow != 0) {
break; }}
if (Overflow == 0) {
exp_xy = norm_l(s);
if (exp_xy<=0)
xy = round(L_shr (s, -exp_xy));
else
xy = round(L_shl (s, exp_xy)); }
mutex_lock(mutex_name);
…
Global variable
int Sim_Time = 0;
Sim_Time += TB();
Sim_Time += TB();
Sim_Time += TB();
Sim_Time += TB();
Sim_Time += TB();
Sim_Time += TB();
wait included
TB() is a function of
# of binary instructions
type of instructions
# of cache misses
frequency
…
even
data dependencies
Sim_Time += TSYS();
TSYS() is a function of
preemptions
conflicts in the bus…
Design Verification & Performance Analysis
25
26. June 8, 2022
Native Simulation
▪ Performance Analysis of ROScpp components
▪ Native simulation of C++ code
▪ Constant time/energy for ROS method calls
▪ Dependent on the CPU
▪ Dependent on the number of nodes and subscribers
▪ Part to be assigned to the component
Design Verification & Performance Analysis
26
27. June 8, 2022
▪ Performance Analysis of ROScpp components
▪ Time/energy for ROS method calls at the component
Design Verification & Performance Analysis
27
28. June 8, 2022
▪ Performance Analysis of ROScpp components
▪ Time/energy for ROS method calls at ROScore
Design Verification & Performance Analysis
28
29. June 8, 2022
▪ Performance Analysis
▪ Estimation Error
Design Verification & Performance Analysis
29
Frequency
(MHz) Estimated Time (ms) Measured Time (ms) Estimation Error (%)
Intel
1600 485.87 582.58 16.60
3000 318.62 461.46 30.95
ARM
600 521.46 904.95 42.38
1500 233.96 398.26 41.25
30. June 8, 2022
▪ Performance Analysis
▪ Error reduction
▪ Improvement increases as the ROS percentage increases
Design Verification & Performance Analysis
30
Frequency
(MHz)
Total Application
time (ms)
%ROS
code
Impact of ROS
estimation error (%)
Impact of no
estimation Improvement
Intel
1600 22,875 2.55 0.42 2.55 83.5%
3000 22,018 2.10 0.65 2.10 69.0%
ARM
600 213,446 0.42 0.18 0.42 57.1%
1500 87,688 0.45 0.19 0.45 57.8%
31. June 8, 2022
▪ Multi-Level Simulation & Performance Analysis
▪ Drone models
Design Verification & Performance Analysis
Abstraction Levels for drone models
Level Drone model Physical model ROS infrastructure
FN Functional No No
FY Functional No Yes
AY Autopilot Yes Yes
AM Autopilot Electro-Mechanical Yes
31
32. June 8, 2022
▪ Drone and robot models
▪ Components in the S3D ‘ModelLibrary’
▪ Instantiation in the ‘VerificationView’
▪ Selection for synthesis of the executable model
Design Verification & Performance Analysis
32
34. June 8, 2022
▪ Multi-Level Simulation & Performance Analysis
▪ Simulation Infrastructure
▪ General architecture
▪ Real-Time (RT) simulation- simulation time = simulated time (SnT = SdT)
▪ As Fast As Possible (AFAP) simulation (SnT as greater as possible than SdT)
Design Verification & Performance Analysis
34
35. June 8, 2022
▪ Multi-Level Simulation & Performance Analysis
▪ Simulation Infrastructure
▪ Functional drone modeling
▪ Without ROS (FN)
▪ Any C++ and ROScpp models (MN + MC + FC + FD)
Design Verification & Performance Analysis
35
36. June 8, 2022
▪ Multi-Level Simulation & Performance Analysis
▪ Simulation Infrastructure
▪ Functional drone modeling
▪ With ROS (FY)
▪ Any C++ and ROScpp models (MN + MC + FC + FD)
Design Verification & Performance Analysis
36
37. June 8, 2022
▪ Multi-Level Simulation & Performance Analysis
▪ Simulation Infrastructure
▪ Realistic drone modeling (Autopilot + Physics)
▪ With ROS (AY + AM)
▪ Any C++ and ROScpp models (MN + MC + FC + FD)
▪ With or without 3D Graphics
Design Verification & Performance Analysis
37
38. June 8, 2022
▪ Multi-Level Simulation
▪ Impact of an increasing number of host CPUs (8 drones)
Simulation Results
38
39. June 8, 2022
▪ Multi-Level Simulation
▪ Impact of an increasing number or realistic vs functional drones
Simulation Results
150
80
39
40. June 8, 2022
Simulation Results
▪ Multi-Level Simulation & Performance Analysis
▪ Impact of an increasing number of realistic drones
40
41. June 8, 2022
▪ Real-Time Simulation in seconds
▪ Impact of an increasing number of realistic drones
Simulation Results
41
42. June 8, 2022
Simulation Results
▪ Real-Time Simulation in % of CPU usage
▪ Impact of an increasing number of realistic drones
42
43. June 8, 2022
Conclusions
▪ Services based on CPS demand new design methods and tools
▪ e.g. Drone-based Services
▪ Close interaction between the physical world and the digital electronics
▪ Model-Driven System Design is a powerful candidate
▪ HiL & SiL are not enough
▪ Model in the loop (MiL) is required
▪ Extension of the classical V-Cycle
▪ Multi-Level Simulation is key in designing drone-based services
▪ As Fast As Possible vs Real-Time
▪ Drones are just pieces inside a complex, distributed functionality
▪ S3D is a valid approach towards MDD of drone-based services
43
44. June 8, 2022
Acknowledgement
▪ Last Research Results from many people
▪ Hector Posadas
▪ Javier Merino
▪ Raul Gonzalez
▪ Jose Maria Gandara
▪ … and the rest of the Microelectronics Engineering Group
44