This document discusses ongoing research projects related to collaborative sensing and heterogeneous networking leveraging vehicular fleets. Specifically, it discusses:
1) How increased cluster density of vehicles improves overall data rates and reduces variability in individual user rates.
2) Modeling what collaborative sensing systems can "see" or be aware of in obstructed environments and how coverage benefits scale with increased penetration of collaborative vehicles.
3) Developing optimal information sharing policies to maximize situational awareness for autonomous nodes in resource-constrained network environments.
Collaborative Sensing and Heterogeneous Networking Leveraging Vehicular Fleets
1. 1
Collaborative Sensing and Heterogeneous Networking
Leveraging Vehicular Fleets
Ongoing Projects
Gustavo de Veciana
Department of Electrical and Computer Engineering,
UT Austin
2. 2
Heterogeneous Networking Leveraging Vehicular Fleets
Gustavo de Veciana, Pablo Caballero and Saadallah Kassir
Department of Electrical and Computer Engineering, UT Austin
Paparao Palacharla, Xi Wang and Nannan Wang
Fujitsu
3. 3
Revisiting Heterogenous Vehicular Networks
V2V V2I
V2V based VANETs (Vehicular
Ad-Hoc Networks)
V2I connectivity to Base
Stations/ RSUs (Road Side
Units)
Internet Connectivity / Infotainment
Clustering
Collaborative Sensing / Mapping
Mobile Computing
+
4. 4
Blocking vehicleMultihomed cluster V2V links
Leveraging V2V connectivity
Overcome challenge of delivering
high bandwidth connectivity
to next generation vehicular
based users
6. 6
• Cluster based opportunism
– Cluster leverages best infrastructure link
– Better rate to vehicles as well as to traditional mobile devices
• Multihoming à Perform load balancing across cells
– Spreading the effect of cell congestion among neighboring cells
– Lower variability in the per user rates
What can one expect from such an approach?
10. 10
Updates
Gustavo de Veciana and Jean A. Rahal
Department of Electrical and Computer Engineering, UT Austin
Takayuki Shimizu and Hongsheng Lu
Toyota
Networked Collaborative Sensing and Distributed
Situational Awareness
11. 11
• Modeling and quantifying what collaborative sensing systems will be
able to “see” in an obstructed/dynamic/stochastic environments.
• Investigating information sharing in collaborative systems to optimize
what autonomous nodes “know” (situational awareness) in resource
constrained network environments.
Past and Ongoing Research Thrusts
12. 12
• Automated driving (or other platforms, e.g., robotic
fleets) require joint perception of the local environment
• Single vehicle’s sensing has limited range and is subject
to occlusions, both of which limit its visibility
• Collaborative sensing can overcome these problems,
but has stringent requirements on communication
Coverage benefits vs. communication costs of collaborative sensing?
How do they scale with penetration of collaborative vehicles?
What can you (we) see?
13. 13
Classical Art Gallery Problems
• computational geometry
Stochastic Art Gallery Problem
• Sensing in the wild
• Stochastic geometry
What you can “see”: Stochastic Art Gallery Problem
16. 16
Coverage Scaling with Penetration
Take-aways:
Collaborative sensing
greatly improves
reliability, e.g., over 80%
reliability at 20%
penetration ratioPerformance
without
collaboration
Benefits of Collaborative Sensing:
Analytical approximation for multilane highway
17. 17
From “raw” sensor data to situational-awareness
• Previous thrust examined the availability sensing data,
i.e., inputs collaborative sensing setting.
• Main goal, however, is to understand how autonomous
nodes should share information in order to optimize
what they can “know” about a dynamic environment.
Situational awareness = timely object recognition/classification
and tracking
What can we “know” ?
18. 18
• What are appropriate metrics for situational awareness?
– Age of information, how long since I got an update?
– Tracking -- Traditional mean square error – tracking
– Object detection – true positive vs false positive
• What are good information sharing policies for real-time distributed
situational awareness on resource constrained environments?
– Context aware, i.e., what is most important “spatially” relevant
– Fairness and/or distributed situational consistency
– New classes of utility maximization frameworks
We are focusing on the following questions:
19. 19
Optimal sampling strategies to minimize remote site MSE
subject to constraint on average update rate !
((#$, &$), $ = ), *, … )W(t)
Sharing time stamped samples
of observed process Estimating remote processObserving random process
End to end
Delays IID
Yi i=1,2,..
20. 20
• Parameterized by ! > 0.
• The next sample to remote site at time "#$% given by
• It is optimal to
1. Wait until the previous sample is received
2. And (only) then send an update if the change in the process exceeds !
&) |)* − ),-
| ≥ ! [1] /) |)* − ),-
| ≤ ! [1]
Optimal Constrained Sampling Strategy
21. 21
Theorem[1]: There exists a ! ≥ 0 such that the sampling policy
is an optimal solution of the MSE optimization problem, and the optimal ! is
determined by solving:
$ max !, )*
+
= -+
. max
1
0123
,
$ max(!+, )*
5
)
2-+!
and the optimal MSE is found to be:
89:;<= =
$ max(!+, )*
5
)
6. $ max !, )*
+ + -+
. $[A]
1 ”Sampling of the Weiner Process for Remote estimation over a Queue”, Y. Sun, Y. Polyansky,
and E. Uysal-Biyikoglu” ISIT 2017. In perparation for Submission.
Optimal Information Sharing Strategy
22. 22
Increased process heterogeneity
Optimizing Rate Allocation for Fair Situational Awareness
3 Sensors Sampling & Broadcasting Heterogeneous Observed Processes
Increased
delay
Rate allocation across users MSE allocation across users
23. 23
• We have been exploring how to best orchestrate information sharing in
distributed situational awareness settings.
• Aim is understand as rigorously as possible simple fundamental models
capturing salient features of the problem.
• Expectation is to get insights that will guide better modeling of
communication constrained information sharing, e.g., higher rates required
to share information which exhibits higher variability over end to end latency
time scales.
Summary Wrapup