An Advanced Extravehicular Activity Robot Assistant

To move beyond teleoperated control of robot manipulators for ISS assembly and maintenance applications, robot controllers that support interactive learning will be required. It is not possible for ground personnel to anticipate all scenarios in which robot assistants in space will be involved and pre-program their response. It will also be increasingly incumbent on the robot to ascertain the intent of the user and be able to learn from user actions to improve its performance. During Phase I, we will implement an unsupervised learning method for selected portions of a manipulator controller. During Phase II we will integrate both supervised and unsupervised learning into an integrated demonstration in which the robot serves as a cooperative assistant. The supervised learning would take place using instrumented gloves or conventional teleoperator controls. The user would bootstrap the robot into operation by demonstrating some portion of a task, after which the robot would improve its performance as it goes. Of course, the operator could intervene at any time to re-program the robot if it becomes stuck. During Phase II, we propose to implement and interface the learning system to the Robotics Research Corp. manipulators at the JSC Cooperative Manipulation Testbed.