Learning Robots Laboratory

The Learning Robots Laboratory at Carnegie Mellon University focuses on combining perceptual reasoning, learning and planning for autonomous mobile robots. In order to be successful, a mobile robot has to be able to deal with incomplete and incorrect state descriptions caused by limited and noisy sensors. It must be able to learn from a limted number of examples, because a robot can not execute millions of trial before getting it right. A robot must robustly interact with the environment both for efficient exploration and to exploit the information it has learned. The planning for the robot must take into account the robots expected performance characteristics and its limited computation.

The ultimate goal of the research undertaken in the laboratory is to create a robot that robustly operates in an office environment for long periods of time, continuously improving its performance through learning. Such a robot must be capable of an autonomous existence during which time its world knowledge is continuously refined through experience and teachings.

It is all too easy to assume that a learning agent has unrealistic initial capabilities such as a prepared environment map or perfect knowledge of its actions. To ground our research, up until 1993 we used a Heath/Zenith Hero 2000 robot (named simply ``Hero''), a commercial wheeled mobile manipulator with a two finger hand, as a testbed on which success or failure is judged. Since 1993, look at Xavier for our latest research.

Summary of Current Research

Summary of Research prior to 1993

Last Updated: 28Jan96 josullvn+@cs.cmu.edu