V. Papudesi and M. Huber (USA)
Reinforcement Learning, Human-Robot Interaction
The automation of various aspects of life through robotics is a promising and useful mechanism to the general end user. Robots are required to accept human guidance, and in its absence, have to operate autonomously while ensuring safety and optimality. This paper presents an approach to variable autonomy that extends reinforcement learning with the capability of integrating user guidance at varying levels of abstraction into its control policies. This permits the modification of robot behavior based on the preferences of the user and faster policy acquisition. User commands are filtered to satisfy a priori constraints and task requirements. The applicability of the approach is illustrated with its operation in a task of navigation.
Important Links:
Go Back
