Cooperative Networked Control of Dynamical PeertoPeer Vehicle Systems

1
Cooperative Networked Control of Dynamical
Peer-to-Peer Vehicle Systems
Geir E. Dullerud University of Illinois UIUC,
MIT, Stanford MURI Interim Meeting January 24,
2003 Sponsored by DDRE and DARPA/AFOSR Program
managers Lt Col Sharon Heise and Dr Belinda King
2
Schedule

• Chair Dullerud
• 830 Opening Remarks. Sharon Heise and Belinda
  King
• 840 Overview and Introduction. Geir Dullerud
• 915 Robotic Vehicle Systems. Eric Feron
• 1000 Break
• Chair Feron
• 1015 Control Subject to Limited Information.
  Sanjay Lall
• 1100 Wireless Networks and Convergence. PR
  Kumar
• 1145 Lunch
• Chair Lall
• 1245 Modeling and Analysis of Complex
  Computational Systems Nancy
  Lynch
• 1330 Overview Summary. Geir Dullerud

3
Schedule (cont)

• Chair Lall
• 1340 Distributed Algorithms for Mobile Sensing
  Networks.
• Francesco Bullo
• 1400 Cooperative Path Planning for Autonomous
  Teams.
• Emilio Frazzoli
• 1420 Verifying Distributed Systems
  Asynchronously.
• Mahesh Viswanathan
• 1440 Management and Control of Mobile Ad-Hoc
  Networks.
• Eytan Modiano
• 1500 Government Caucus
• 1530 Conclusion

4
Major Motivating Applications

• Cooperative vehicle teams
• Military targets, refueling, safe-zones also
  symmetric
• Civilian security, firefighting

5
Goals

• Establish theory, scalable control algorithms and
  protocols for cooperative networked control of
  vehicle systems
• Performance and correctness verifiable with
  robustness to
• external uncertainty
• malicious attack
• rapidly evolving objectives

6
Expected Deliverables
Include

• Robustness theory for distributed systems
• Scalable algorithms for verification of
  cooperative vehicle systems
• Ad hoc communications protocols for networked
  control
• Allocation algorithms based on spatial geometry

• Information management in large-scale feedback
  systems
• Languages for vehicle negotiation and interaction
• Randomized and real algebraic techniques for
  related combinatorial problems
• Examples and proof-of-concept implementations on
  testbeds

7
Projected Impact

• Theory
• new modeling paradigms
• theoretical tools
• Computation
• algorithms and heuristics
• software
• examples and implementation
• Applications
• collaborations with DOD labs
• industrial transitions

8
Research Approach

• Connections between verification methods of
  control and computing
• Exploit dynamical and spatial structure
• Communications for control
• Control for communications
• Leverage recent computation and analysis tools

9
Team
MIT
Stanford
Illinois

• Francesco Bullo
• Geir Dullerud
• Emilio Frazzoli
• P.R. Kumar
• Daniel Liberzon
• Bruce Reznick
• Mahesh Viswanathan

• Jinane Abounadi
• Eric Feron
• Nancy Lynch
• Sanjoy Mitter
• Eytan Modiano

• Sanjay Lall
• John Mitchell

10
Team controls
MIT
Stanford
Illinois

• Francesco Bullo
• Geir Dullerud
• Emilio Frazzoli
• P.R. Kumar
• Daniel Liberzon
• Bruce Reznick
• Mahesh Viswanathan

• Jinane Abounadi
• Eric Feron
• Nancy Lynch
• Sanjoy Mitter
• Eytan Modiano

• Sanjay Lall
• John Mitchell

11
Team computing
MIT
Stanford
Illinois

• Francesco Bullo
• Geir Dullerud
• Emilio Frazzoli
• P.R. Kumar
• Daniel Liberzon
• Bruce Reznick
• Mahesh Viswanathan

• Jinane Abounadi
• Eric Feron
• Nancy Lynch
• Sanjoy Mitter
• Eytan Modiano

• Sanjay Lall
• John Mitchell

12
Teamcommunications
MIT
Stanford
Illinois

• Francesco Bullo
• Geir Dullerud
• Emilio Frazzoli
• P.R. Kumar
• Daniel Liberzon
• Bruce Reznick
• Mahesh Viswanathan

• Jinane Abounadi
• Eric Feron
• Nancy Lynch
• Sanjoy Mitter
• Eytan Modiano

• Sanjay Lall
• John Mitchell

13
Teamdisciplines
MIT
Stanford
Illinois

• Francesco Bullo
• Geir Dullerud
• Emilio Frazzoli
• P.R. Kumar
• Daniel Liberzon
• Bruce Reznick
• Mahesh Viswanathan

• Jinane Abounadi
• Eric Feron
• Nancy Lynch
• Sanjoy Mitter
• Eytan Modiano

• Sanjay Lall
• John Mitchell

14
Recent Project Interactions

• MURI Workshop, Oct 4, 2002 UIUC
• UIUC bimonthly seminars
• Other interactions
• AFRL/GERC, Eglin AFB
• AFRL, Wright-Patterson AFB
• Honeywell Labs, Minneapolis

15
Major Research Issues

• Scalability and computation
• Uncertainty
• dynamics
• attack and failure
• Ad hoc network communication
• Modeling and specification
• Language and Information

16
Cooperative Vehicle Systems

• Scenarios
• Geographic
• mapping
• coverage
• search and rescue
• strategic reconnaissance
• Strategic
• coordinated attack
• pursuer-evader
• enemy containment
• formations

• Communications
• disruption
• mobile Internet
• distributed data bases
• Other team games
• Leader-based versions
• vehicle dynamics
• communications
• uncertainty

Idealized problems
17
Illinois Testbeds

• IT Convergence (Kumar)
• RC cars (currently 15)
• Linux PCs, 802.11 communications
• wireless protocols and real time control

• HOTDEC (Dullerud)
• Hovercraft
• Linux-based
• Bluetooth and Internet
• 8-10 planned (currently 2)
• many cooperative scenarios

18
Illinois Testbeds
Multi-Rover Network (Bullo)
AeroNet (Frazzoli)

• Outdoor rover network (10 vehicles)
• GPS positional sensing
• Coverage problems

• Under development
• RC airplanes (2-4 planned)
• Electric power
• GPS and inertial guidance

19
Stanford Testbed

• SWARM (Lall)
• table-based robotic vehicles experiment
• wireless and wired media
• 6-8 vehicles
• multiple scenarios
• search and rescue
• evade/capture
• team games
• mapping
• under development

20
Distributed Vehicles Testbeds
complexity
AeroNet Frazzoli
Rovers Bullo
HOTDEC Dullerud
IT Convergence Kumar
SWARM Lall
implementation
21
Approach
DISTRIBUTED
DISTRIBUTED
CONTROL
COMPUTING
Computing And Verification
Control and Dynamics
Uncertainty
Automated
analysis
verification
Dynamical
Semantics
systems
and language
Semidefinite
Randomized
programming
algorithms
Semialgebraic
Combinatorial
methods
optimization
Information theory
Network topology
and feedback
and protocols
Wireless Communications
Information Theory
22
Technical Themes
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
CONTROL
COMPUTING
CONTROL
Automated
Uncertainty
ROBUSTNESS
verification
analysis
Semantics
Dynamical
and language
systems
MODELS
INTERFACES
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Combinatorial
Semialgebraic
optimization
methods
Network topology
Information theory
GEOMETRY
and protocols
and feedback
23
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
Dullerud
Mitchell
CONTROL
COMPUTING
CONTROL
Lall
Automated
Uncertainty
Lynch
ROBUSTNESS
verification
analysis
Frazzoli
Semantics
Dynamical
Liberzon
and language
systems
MODELS
INTERFACES
Feron
Viswanathan
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Bullo
Combinatorial
Semialgebraic
Kumar
optimization
methods
Reznick
Network topology
Information theory
GEOMETRY
and protocols
and feedback
Abounadi
Mitter
Modiano
24
Research Areas
Control Information Theory
Computing Verification
Robotic Vehicles
Communications
25
Robotic Vehicle Systems
Control Information Theory
Computing Verification
Robotic Vehicles
Bullo Feron Frazzoli
Communications
26
Coverage and Sensing Networks
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
CONTROL
COMPUTING
CONTROL
Automated
Uncertainty
ROBUSTNESS
verification
analysis
Semantics
Dynamical
and language
systems
MODELS
INTERFACES
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Combinatorial
Semialgebraic
optimization
methods
Network topology
Information theory
GEOMETRY
and protocols
and feedback
27
Coverage and Sensing Networks

• Optimal partitioning of space
• regions of domination
• information-based sensing
• predicting adversaries
• Vehicle agents
• dynamical
• controllable
• Large-scale distributed protocols

28
Voronoi Tesselations
initial
centroidal
execution
Criteria
probabilistic
worst-case
29
Coverage and Sensing

• dynamic vehicle constraints (e.g., aircraft)
• non-convex unknown environments
• non-isotropic sensors
• vehicle diversity
• other cooperative tasks e.g.
• exploration and map building
• target identification
• interface with estimation/detection,
  communications, formation control
• nonstationary rapidly changing objectives
• Work by Bullo

30
Multi-Vehicle Motion Planning
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
CONTROL
COMPUTING
CONTROL
Automated
Uncertainty
ROBUSTNESS
verification
analysis
Semantics
Dynamical
and language
systems
MODELS
INTERFACES
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Combinatorial
Semialgebraic
optimization
methods
Network topology
Information theory
GEOMETRY
and protocols
and feedback
31
Multi-Vehicle Motion Planning

• Can we develop an abstract framework?
• generation and execution in real time
• compatible with actual dynamics
• distributed communication and negotiation
• New automaton or Language based approach
• restrict to maneuvers or motion primitives
• exploit symmetry in vehicle design

32
Symmetry

• A fundamental geometric property of vehicles
• Symmetry dynamics invariant under the group
  action ? H ? X ? X. Namely

33
Maneuver Automaton
Classes of trajectory primitives (trim
maneuver)

• construct a maneuver library with finite
  primitives
• a new language via sequencing motion primitives.

34
Automaton Model
Current
Longer Term

• Multivehicle
• language generalizations
• distributed structure
• collision avoidance
• synchronization
• Robustness
• Randomized algorithms
• Noncooperative scenarios
• Information theory

• Controllability results
• Optimal control
• feasibility in constant time
• mixed integer LP
• Some robustness results
• Considered motion planning with obstacles
• See talks by Feron, Frazzoli

35
Computing and Verification
Control Information Theory
Computing Verification
Robotic Vehicles
Lynch Mitchell Viswanathan
Communications
36
Verification for Cooperative Systems
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
CONTROL
COMPUTING
CONTROL
Automated
Uncertainty
ROBUSTNESS
verification
analysis
Semantics
Dynamical
and language
systems
MODELS
INTERFACES
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Combinatorial
Semialgebraic
optimization
methods
Network topology
Information theory
GEOMETRY
and protocols
and feedback
37
Verification for Cooperative Systems
Specifications mixed

• diff. equations
• stability and asymptotic performance
• norm-based performance

• algorithms
• liveness
• safety
• fairness

Need deep connection made between verification in
disciplines of computing and controls.

Challenges include

• modeling
• large-scale

• tractability
• robustness

38
Model Properties and Paradigms
Abstraction and Decomposition
Example set intersection

• Introduce nondeterminism reduce complexity
  always present in physical systems, i.e.,
  robustness
• Break up verification task tricky proof of one
  part requires assumptions about the others
  correctness
• Combination
• Approximate correctness accept probabilistic
  answers

39
Example Input-Output Automata
Probabilistic hybrid IOA
Hybrid IOA
Probabilistic timed IOA
Probabilistic IOA
Timed IOA
IOA (discrete)

• Theory of IOA for purely discrete systems, timed
  version
• Recent framework for hybrid systems still in
  progress
• supports abstraction and composition
• Probabilistic systems for reasoning important
• Work by Lynch

40
Verification Directions

• Modeling
• abstraction
• probabilistic discrete
• hybrid
• probabilistic
• robust control models
• Implementation
• semidefinite programming
• randomized algorithms
• bounded error
• zero error
• complexity analysis
• See talk by Lynch
• Work by Viswanathan

• Methods
• computer science
• model checking
• theorem provers
• compositional reasoning
• property testing
• controls
• robust control
• Lyapunov approaches
• Lie algebraic
• geometric methods
• comparisons
• can methods effectively cross areas?
• what types of problems can be handled?

41
Communications
Control Information Theory
Computing Verification
Robotic Vehicles
Abounadi Kumar Modiano
Communications
42
Wireless Communications in Cooperative Mobile
Networks
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
CONTROL
COMPUTING
CONTROL
Automated
Uncertainty
ROBUSTNESS
verification
analysis
Semantics
Dynamical
and language
systems
MODELS
INTERFACES
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Combinatorial
Semialgebraic
optimization
methods
Network topology
Information theory
GEOMETRY
and protocols
and feedback
43
Wireless Communications in Cooperative Mobile
Networks
Goals
Environment

• distributed control
• command and control
• sensing networks
• formation flight
• data networks
• mobile Internet
• databases

• no fixed network infrastructure
• cell stations, backbone, et cetera
• rapidly changing connectivity
• route changes
• time-varying link quality
• error rates, data rates
• limited and varied power,energy, mobility
• mix of vehicle capabilities

44
Wireless Cooperative Networks

• We want transmission that is
• predictable latency control
• bandwidth guarantees, capacity data transfer and
  control
• Technical challenges
• routing protocols rapidly changes topology
• power and energy control unicast, multicase
• ad-hoc networks architectures media access
• robustness and reliability QoS, failure
• See talk by Kumar
• Work by Abounadi and Modiano

45
Wireless Cooperative Networks
How should we operate network?

• Multi-hop Transport
• complete decoding at each node
• treat interference as noise

Interference noise
Interference noise
Interference noise
Interference noise
packets

• Cooperative Relaying

• coherent multistage relaying w/ interference
  cancellation (COMSRIC)
• upstream nodes cooperate to advance packet to
  next node

Example interference or noise

• want soft-signal data
• decode loud signal
• then decode soft signal

46
Control and Information Theory
Control Information Theory
Computing Verification
Robotic Vehicles
Dullerud Lall Liberzon Mitter Reznick
Communications
47
Decentralized Control
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
CONTROL
COMPUTING
CONTROL
Automated
Uncertainty
ROBUSTNESS
verification
analysis
Semantics
Dynamical
and language
systems
MODELS
INTERFACES
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Combinatorial
Semialgebraic
optimization
methods
Network topology
Information theory
GEOMETRY
and protocols
and feedback
48
Decentralized Control
Plant

• Global criteria with local control
• global design, local implementation
• local design with local implementation
• Recent results on distributed case
• multidimensional systems theory
• diagonal dominance
• See talk by Lall current advances

49
Information and Control
COOPERATIVE
DISTRIBUTED
DISTRIBUTED
NETWORKED
CONTROL
COMPUTING
CONTROL
Automated
Uncertainty
ROBUSTNESS
verification
analysis
Semantics
Dynamical
and language
systems
MODELS
INTERFACES
Randomized
Semidefinite
algorithms
programming
ALGORITHMS
Combinatorial
Semialgebraic
optimization
methods
Network topology
Information theory
GEOMETRY
and protocols
and feedback
50
Information and Control
Confluence of information theory and control
theory

• Information in dynamic setting
• what is a definition of information in this
  setting?
• how much information is needed to perform a given
  control task?
• Control with limited information
• can we perform a given control task with given
  information constraints?
• what control tasks can be performed with given
  information constraints?

51
Quantized Control
q(x)
x
PLANT
QUANTIZER
x
q(x)
CONTROLLER
Dynamic Quantization
52
Coding and Control
code decode
code decode
noisy delay channel

• limited information pattern
• noise and nonstationary delays
• multiple agents with shared channels

53
Information Theory and Cooperative Control

• Develop a dynamical view of information theory
  fundamental limits of performance
• stabilization and performance
• time-scale issues
• may provide insight for pure distributed comms
• Multiple distributed agents
• Minimum attention control
• Stochastic setting
• Work by Liberzon, Mitter

54
Computational and Mathematical Methods

• Randomized algorithms
• model checking
• property testing
• path planning
• Semidefinite Programming
• control problems
• relaxations of combinatorial problems
• Real Algebraic Techniques
• theory
• computation
• semialgebraic set
• Work by Reznick see also talk by Lall

55
Research Areas Composite Talks
Control Information Theory
Computing Verification
Lall
Lynch
Robotic Vehicles
Feron
Kumar
Communications