MultiRobot Lab Mobile Autonomous Robot Software (MARS)

Project Objectives

Develop complete, effective and scalable software for autonomous robot teams. Demonstrate robot teams with integrated action, perception, reasoning, communication and cooperative strategies that solve complex multiagent tasks.

Approach

Autonomous robots face many complexities in the real world, in particular: uncertainty about the effects of their actions, large numbers of potential state features and coexistence with multiple cooperative and potentially adversarial robots. In these complex tasks, it is impossible to sufficiently model and identify all the relevant world features necessary for effective goal achievement beforehand. Instead, autonomous robots must discover this knowledge themselves as they interact with their environment.

This research targets the fundamental challenges in the successful deployment autonomous learning cooperating robot teams:

• Bridging the gap between low-level and high-level control: We utilize a multi-layered robot control architecture. While it seems clear that a multi-layered solution is the correct way to address the representational and functional gap between low-level and high-level control, important, non-trivial challenges remain. It is not yet clear, for instance, how to break a particular task into appropriate layers. This research investigates learning algorithms for automatically determining appropriate control layers for particular tasks.

• Reusable subproblems: Any complex task can be decomposed into simpler subtasks with reusable solutions. The ability to identify and effectively reuse subproblem solutions is of crucial importance in the successful deployment of autonomous robots. One thrust of this work is control learning in subproblems for reuse both in individual tasks and for transfer between robots in multiagent robotic tasks.

• Real-time adversary modeling: In military environments robots will face other agents that act according to strategies initially unknown to our robots. By observing and modeling the behavior of opposing agents in real-time, robot teams can act more effectively. We are developing a real-time learning algorithm to recognize a foreign robot's strategic model and to adapt our own strategies accordingly.

• Scalable command and control for large numbers of autonomous robots: The question of how to coordinate the activities of autonomous robot teams is still rather open, especially for large numbers of agents. In this work we are developing a scalable architecture suitable for controlling tens or hundreds of autonomous robots.

Recent Accomplishments

• Acquired a mobility platform (Personal Robotic's Cye robot) for evaluation for use in this project.

• Prototyped a computing and sensor platform for use as a controller for the Cye. The system includes:
  • PC-104Plus motherboard (Pentium 133),
  • 2GB laptop hard disk,
  • Imagenation color frame grabber,
  • miniature color camera.

Plan

Over the next year our research efforts will focus on on:

• Design and prototyping of a multi-layered multirobot control architecture including the integration of planning and reactive control.
• Adversary modeling: recognize the behavior of another robot in simulation and on a mobile robot.
• Reusable subproblems: demonstrate the transfer of a learned solution from simulation to mobile robots. Merge and adapt multiple simple subproblems to solve more complex problems.

• Integrate computing and mobility platforms.
• Develop Java-based software API for controlling the mobility platform.
• Evaluate the mobility platform and build multiple copies for multirobot research.
• Investigate alternative wireless communication technologies.

Technology Transition

• All software will continue to be available on the net. Our TeamBots software is already available online and is used by a number of researchers world-wide.