Adaptive Autonomous Robot Teams for Situational Awareness

Gaurav S. Sukhatme, Maja J. Mataric,

Andrew Howard, Ashley Tews

Robotics Research LaboratoryUniversity of Southern California

http://robotics.usc.edu/projects/mars

Task Summary• Outdoor simulation

• Cooperative outdoor localization

• Semantic representations

• Stealthy behaviors

• Path-referenced perception

• Human-robot interface

• Integration

Simulation• Planned Extensions

– 3D simulation for outdoor terrain.

– Incorporate USC helicopter and UPenn blimp

• Recent Progress: Player – generic driver/interface model – IMU/GPS drivers, laser and

visual fiducial detectors– Supports multiple architectures

(e.g. Solaris, iPAQ)

Cooperative Outdoor Localization

• Plans– Extend existing localization

algorithms to outdoor environments.

– Implement outdoor localization in the presence of partial GPS.

– Validate through outdoor experiments with small teams (4 ground robots).

Semantic Representation and Activity Recognition

• Semantic mark-up of maps with following attributes: – elevation, terrain type and traversability, foliage and

coverage type, and impact on communications.

• Integrate activity/motion detection algorithms to locate people in the environment.

• Demonstrate semantic markup using ground robots at USC.

Variable Autonomy and Stealth

• Develop and implement behaviors for variable autonomy incorporating operator feedback using gestures

• Develop and implement a new “stealthy patrolling” behavior by integrating visibility constraints into current patrolling algorithms

• Adapt and tune above behaviors using reinforcement learning to improve performance

Path-referenced Perception and Behaviors

• Develop path-referenced perception and behaviors, which allow recall of behavioral strategy relative to priors paths taken in the mission

• Integrate path-referencing which allows robots to query each other for relative locations of semantic mark-ups

Integration• Plans:

– Demonstrations at USC of cooperative localization (laser based with IMU and GPS) using ground robots and USC helicopter.

– Demonstration at USC of activity detection, semantic markup of terrain and stealthy traverses.

– Support joint demonstration with ground robots.

Other Progress to Date

• Contract in place: Nov 4, 2002• New Personnel: Nathan Koenig (student),

Emil Birgesson (visitor from Sweden), Ian Kelly (100%), Stefan Hrabar (50%) both as project specialists

• New Robots: – 4 Pioneer 2ATs (Intel gift)– Control boards and IMU under test