A method is proposed for enabling human operators to teach anthropomorphic robots, in the field, how to perform new complex tasks. Building upon existing inverse kinematics, rule-based control, and neural network learning technology, the training paradigm enhances robot capabilities through operator supervision. Rules provide a modular and hierarchical way to specify plans of action, and neural networks provide a context-dependent form of skill memory. The proposed innovation enables the online construction of rule-based plans through a verbal dialogue between operator and robot. Additionally, it uses verbal, visual, and haptic cues such as spoken words, hand gestures, and the pushing of buttons and joysticks, along with a variation of feedback-error learning, to shape robot behavior and thereby augment the nominal movement provided by rule-based plans. In Phase I, this integrated training paradigm is applied to a simulated humanoid robot under pick-and-place test scenarios, including the assembly of a truss. It is evaluated for its ability to provide intuitive, fast, incremental, and robust robot skill acquisition. The anticipated show-and-tell approach to training will give future NASA robots an unending ability to learn, and NASA astronauts the ability to customize robot behavior for both routine tasks and unexpected situations. POTENTIAL COMMERCIAL APPLICATIONS Successful completion of the proposed effort will result in a trainable robot control system with great commercial potential. The incremental nature of the training algorithm permits an operator with limited mobility to build complex tasks. At no time is the operator expected to provide a complete, precise example of task performance. Robot competency builds over time, course to fine, given symbolic task plans and error-driven corrective cues. Not only can suited NASA astronauts with restricted movement benefit from this approach, but also persons with disabilities. With this technology, the disabled will be able to tailor to their personal needs the behavior of robotic assistive devices, providing a level of access and a sense of freedom otherwise difficult to attain. More generally, as anthropomorphic service robots begin to appear in consumer markets, intuitive man-machine interfaces will promote acceptance of these devices. Furthermore, smart robots capable of learning will ultimately find their way into entertainment and educational products. The proposed effort addresses these opportunities.
