MultiRobot Lab Mobile Autonomous Robot Software (MARS)

Project Objectives

We remain focused on our objectives:

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

Project Website

More details on our project and copies of our presentations are available on the project website: http://www.cs.cmu.edu/~multirobotlab/MARS

Approach

Over the last year, our research has focused on creating individually skilled autonomous robots and successful multirobot teams. The key challenges include:
• Localization
• Manipulation and navigation
• Graceful degradation in the face of reduced knowledge
• Multiagent learning
• Communication and fusion of distributed sensing
• Real-time visual sensing

Localization
We have developed new algorithms for visual landmark-based localization. The approach specifically addresses situations where sensor information does not match estimated position. Our algorithms have been implemented on the Sony Aibo, and ported to the TeamBots environment.

Manipulation and Navigation for Robot Colonization
We pose ``robot colonization'' as a task domain for our multirobot research. In this task, robots are to search for various types of objects in an unmapped environment and transport them to specific locations according to type. Some objects may be of greater value than others, thus requiring dynamic replanning for optimal solutions.

Manipulation and navigation are fundamental skills for this task. Accordingly, we have been developing behaviors to accomplish these skills. Photos of our robots demonstrating components of the colonization task are available in our presentation.

Graceful degradation in the face of reduced knowledge
Robots acting in the real world are often faced with varying degrees of information about their environment. For example, sometimes they know very precisely their location, other times they are lost. It is important that the robots act appropriately in all these situations. Our work in this area has focused on developing behaviors, called ``multi-fidelity behaviors'' that enable the agent to act appropriately regardless of the confidence in their confidence in sensing.

Multiagent learning
Learning enables agents to adapt to their environment and to interact appropriately with external agents (or perhaps opponents). We have been investigating learning algorithms for this domain that speed convergence to effective policies.

Communication and fusion of distributed sensing
We are investigating how robots can communicate about objects in their environment, and fuse this information effectively.

Real-time visual sensing
In order for our robots to react sensibly to the real world, they must be able to quickly sense objects in their environment. We have developed a software system that can process color images and extract regions based on color coding at 30Hz with very low CPU overhead. This software has been downloaded and used successfully by a number of researchers.

Recent Accomplishments

Our progress over the last year is reported in detail in the presentations available on our project website. Here is an overview:
• We have constructed 5 robots as a multiagent testbed for our research.
  • perception: each robot has it's own vision system and wireless communication capabilities.
  • cognition: multi-fidelity behaviors using the TeamBots architecture.
  • action: controlled navigation and manipulation.

• CMVision - applied to a variety of different robotic platforms:
  • Minnow robots.
  • Sony Dog robots.
  • System used for observation of social insects.
  • Observation of dynamic multirobot soccer.

• Learning - new multiagent learning algorithm. Learner has been proven to act rationally in the presence of other learners. Learner effectively empirically converges to mixed optimal policy Behavior Recognition

• Behavior Recognition - extended Hidden Markov Models to the problem of behavior. Recognition applied to observing a single robot.

Plan

Over the next year our research efforts will focus on:
• Continue the research on fusion of vision and communication in multi-robot systems
• Scale to 10 robots fully autonomous and 10 other smaller robots with limited capabilities
• Investigate the control of limited-capability robots by communication from fully-autonomous ones
• Continue the study of robot behavior recognition extending it to multi-robot behavior and behavior learning

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. The following robot designs and software funded under the MARS program have been developed by us and released to the community:
• The Minnow multirobot system.
• The TeamBots software.
• CMVision realtime color vision.

Relevant Publications

• Heterogeneous Multi-Robot Systems, special issue of Autonomous Robots edited by Tucker Balch and Lynne Parker, July 2000.

• "Automated robot behavior recognition applied to robotic soccer", Kwun Han and Manuela Veloso. In Proceedings of the 9th International Symposium of Robotics Research (ISSR'99)}, pages 199--204, Snowbird, Utah, October 1999.

• "Sensor resetting localization for poorly modelled mobile robots", Scott Lenser and Manuela Veloso, In Proceedings of ICRA-2000, the International Conference on Robotics and Automation, April 2000.

• "Vision-servoed localization and behavior-based planning for an autonomous quadruped legged robot", Manuela Veloso, Elly Winner, Scott Lenser, James Bruce, and Tucker Balch. In Proceedings of AIPS-2000, the Fifth International Conference on Artificial Intelligence Planning Systems, Breckenridge, CO, April 2000.

• "Real-Time Color Image Segmentation Using Commodity Hardware", James Bruce and Tucker Balch and Manuela Veloso, Proceedings of WIRE-2000", May 2000.

• "Social Potentials for Scalable Multi-Robot Formations", Proceedings of ICRA-2000, Tucker Balch and Maria Hybinette.

• "Multi-fidelity behaviors: Acting with variable state information", Elly Winner and Manuela Veloso. In Proceedings of AAAI-2000, July 2000.

• "On behavior classification in adversarial environments", Patrick Riley and Manuela Veloso. In Proceedings of AAAI-2000, July 2000, student abstract.

• RoboCup-99: Robot Soccer World Cup III Manuela Veloso, Enrico Pagello, and Hiroaki Kitano, editors. Springer-Verlag Press, Berlin, August 2000.

• "Fast and inexpensive color image segmentation for interactive robots", James Bruce, Tucker Balch, and Manuela Veloso, In Proceedings of IROS-2000, October 2000.

• "Rational learning of mixed equilibria in stochastic games", Michael Bowling and Manuela~M. Veloso. Submitted to NIPS-2000, November 2000.