Authors:
Diego Pardo
and
Cecilio Angulo
Affiliation:
GREC - Knowledge Engineering Research Group, UPC - Technical University of Catalonia, Spain
Keyword(s):
Robot control architecture, Sensorimotor learning, Coordination policy, Reinforcement learning.
Related
Ontology
Subjects/Areas/Topics:
Autonomous Agents
;
Informatics in Control, Automation and Robotics
;
Mobile Robots and Autonomous Systems
;
Modeling, Simulation and Architectures
;
Robotics and Automation
Abstract:
This paper describes a collaborative control scheme that governs the dynamic behavior of an articulated mobile robot with several degrees of freedom (DOF) and redundancies. These types of robots need a high level of coordination between the motors performance to complete their motions. In the employed scheme, the actuators involved in a specific task share information, computing integrated control actions. The control functions are found using a stochastic reinforcement learning technique allowing the robot to automatically generate them based on experiences. This type of control is based on a modularization principle: complex overall behavior is the result of the interaction of individual simple components. Unlike the standard procedures, this approach is not meant to follow a trajectory generated by a planner, instead, the trajectory emerges as a consequence of the collaboration between joints movements while seeking the achievement of a goal. The learning of the sensorimotor coord
ination in a simulated humanoid is presented as a demonstration.
(More)