Autonomous Group Behaviors

1
Autonomous Group Behaviors Dr. Vijay Kumar
University Of Pennsylvania October 15th, 2008
Army Research Laboratories BAE Systems Georgia
Institute of Technology Jet Propulsion
Laboratory Massachusetts Institute of
Technology University of California,
Berkeley University of California,
Merced University of Delaware University of
Maryland University of Michigan University of New
Mexico University of Pennsylvania Vanderbilt
University
Distribution authorized to DoD and DoD
Contractors Only Critical Technology (March
2007). Other requests for this document Shall be
referred to Director, US Army Research
Laboratory, ATTN AMSRD-ARL-DP-P, Adelphi, MD
20783-1197
2
MAST Vision
To develop autonomous, multifunctional,
collaborative ensembles of agile, mobile
microsystems to enhance tactical situational
awareness in urban and complex terrain for small
unit operations.
Autonomy
Versatility
Cooperation
Uncertainty
Complexity
3
Challenges for Autonomy

• Must have multiple, heterogeneous Autonomous
  Mobile, Multifunctional Microsystems (AM3)
  functioning as a single cohesive unit
• Ensemble must be adaptable, responsive to human
  commands and resilient to adversarial conditions
• Need to integrate control, sensing,
  communication, perception, and planning
  methodologies

4
Examples of MAST tasks

• Achieve eyes on target in an urban environment
  without commanding individual assets with poor
  GPS and lossy communications
• Enter into a building, organize into subteams and
  clear the building, and then secure the building
  
• Provide an integrated picture for situational
  awareness to war ?ghters

• Track hostile targets and dynamic threats in 3-D
  environments
• Find victims and IEDs in damaged buildings with
  crawling, climbing, ?ying, or hovering vehicles
• Maintain persistent surveillance of a rural or
  urban environment
• Provide situational awareness of poorly lit,
  uneven, gusty, unstructured environments (caves,
  subterranean pipes)
• Surround perimeters of buildings or compounds for
  perimeter defense

Multi-Vehicle Situational Awareness
Maps State estimates Mutual info
Tasks Plans Controllers
Software
Maps State estimates Mutual info
Tasks Plans Controllers
models of platforms, environment,
resources, algorithms
Group Navigation, Control, Communication
Distributed Sensing and Perception
Fading Shadowing Interference SNR Link State

Fading Shadowing Interference SNR Link State
MAST Environments
5
Three Themes

• 1. Multi-Vehicle Situational Awareness
• 2. Group Navigation and Control with

• Constrained
• Communication
• 3. Distributed Sensing
• and Perception

Multi-Vehicle Situational Awareness
Maps State estimates Mutual info
Tasks Plans Controllers
Software
Maps State estimates Mutual info
Tasks Plans Controllers
models of platforms, environment,
resources, algorithms
Group Navigation, Control, Communication
Distributed Sensing and Perception
Fading Shadowing Interference SNR Link State

Fading Shadowing Interference SNR Link State
MAST Environments
6
Overarching Theme
Biological Inspiration
7
Closely Related to Themes in other Thrusts

• Adaptive, Agile, Mobile Systems for Operation in
  Complex Environment
• Platform navigation and control
• Systems of Microsystems
• MAST Integrated Design Analysis Simulation

8
Related ThemeNavigation and Control

• Abstraction-based control of MAST platforms
• Pappas (Penn), Tanner (UD), Kumar (Penn)
• Model Predictive Navigation in unstructured
  environments
• Tanner (UD), Jadbabaie (Penn), Sastry (Berkeley)
• Navigation Using Spatio-Temporal Gaussian
  Processes
• Sastry (Berkeley), Oh (Merced)
• Millimeter Wave Radar
• East (Michigan)
• HAIR Inertial Navigation, Sensing and Actuation
• Pierce, Najafi (Michigan)

Bugbot Project Bundy (ARL/WMRD), Fearing
(Berkeley), Michael (Penn), Pierce (Mich)
9
1. Multi-Vehicle Situational Awareness

• Active Simultaneous Localization and Mapping
• Christensen (GT), Dellaert (GT), Jadbabaie
  (Penn), Kumar (Penn)
• Decentralized coverage verification and
  cooperative surveillance
• Jadbabaie (Penn), Pappas (Penn), Tomlin
  (Berkeley)
• Autonomous adaptive mobility of heterogeneous
  teams
• Kumar (Penn), Rus (MIT)
• Composition of group behaviors for scouting,
  reconnaissance, and surveillance
• Kumar (Penn), Pappas (Penn)

10
Active Simultaneous Localization and Mapping

• Control for mapping
• Localization for control
• Mapping for localization
• Localization for mapping

Potential Project Fields (ARL/VTD)
11
Decentralized coverage verification and
cooperative surveillance

• Verify coverage in a robot network with no
  location information
• Localize coverage holes
• Find the minimal number of sensors necessary to
  cover the region.

With minimal sensing
In a distributed manner
12
Autonomous adaptive mobility of heterogeneous
teams
MAST
ARO SWARMS MURI
Potential Project Sadler (ARL/CISD), Lindsey
(Penn)
13
Composition of group behaviors for scouting,
reconnaissance, and surveillance
5 helicopters and 25 ground vehicles navigate to
a designated building and surround the site
14
Theme 2 Group Navigation and Control

• Communication-aware exploration
• Mostofi (UNM)
• Simulation environment for the integration of
  communication and navigation
• Baras (Umd), Kumar (Penn), Mostofi (UNM)
• Integration of Communication and Control in MAST
  platforms
• Jadbabaie (Penn), Mostofi (UNM)
• Coordinated control of a small team of
  heterogeneous robots in partially-known
  environments
• Kumar (Penn), Likhachev (Penn)
• Path Planning for Multi-Agent Teams in dynamic
  environments
• Jadbabaie (Penn)
• Avian-based wing morphing
• Hubbard (UMd)

15
MAST Integrated Simulation Environment
Wiki site, repository
16
Decentralized Control for MAST
50
20
Environment degree of difficulty
Number of vehicles
10
5
complexity of coordinated control task
3
State of the art
1. Coordinated Control in unstructured
environments
2. Coordinated control in 2.5D indoor/outdoor
environments
3. Coordinated control in 3D, feature-rich
environments (rubble, caves)
4. Perimeter surveillance and coverage in outdoor
environments
17
Distributed Sensing and Perception

• Distributed Inference
• Dellaert (GT), Durrant-Whyte (Usyd)
• Simultaneous localization and mapping
• Christensen (GT), Daniilidis (Penn)
• Mapping and localization in 3-D, unstructured
  environments
• Christensen (GT), Daniilidis (Penn), Dellaert
  (GT)
• Detecting, tracking and classifying targets
• Oh (Merced), Sastry (Berkeley)
• 4-D structure and semantics for situational
  awareness
• Daniilidis (Penn)

Potential Project OBrien, Young (ARL/CISD)
18
Distributed Perception for MAST
50
106
Number of required nodes
20
105
Number of features
Number of MAST nodes required for situational
awareness
104
10
5
103
complexity of mapping and providing situational
awareness
3
102
State of the art
1. Navigation, mapping, SA in 2D and 2.5D indoor
environments
2. Navigation, mapping, SA in 2D and 2.5D
indoor/outdoor environments
3. Navigation, mapping, SA in 3D, feature-rich
environments (rubble, caves)
4. Perimeter surveillance and coverage in outdoor
environments
19
Model-Based Design, Integration and Validation of
MAST Software
Language
Penn
Mission specification
BAE, GT
Penn
Planning
Controller synthesis
Berkeley
U Md
Model-based code generation
Vanderbilt
End-to-end verification/validation
Penn
Testing and validation with simulation
and experimental platforms
Berkeley, GT, Penn
20
Concluding Remarks

• Group Autonomy requires close coordination with
  other thrusts
• Small size, unstructured environment and
  heterogeneity require unique approaches
• Real-world constraints introduce new challenges
• Complex environments
• Communications
• Stochastic processes
• Noisy sensors