Learning Control Behaviour within the Control Basis Framework

The Control Basis Framework (by Grupen et. al. 1998) is a powerful approach to closed loop control. This project aims at providing a library implementing the Control Basis Framework idea and possibly extending it to concurrent execution. Additional research is planned to investigate how to make machines learn to utilize the control affordances provided by synthesized controllers. Within the Control Basis Framework controllers are assembled from individual parts, like artificial potential fields, sensor and effector transformations, and robot resources. Additionally hierarchical composition of controllers is supported by means of jacobian null-space projection of the gradient steps of subordinate controllers. This allows to reach secondary goals, like posture optimization, obstacle avoidance, etc., given that there exist sufficiently many degrees of freedom to effectively make use of the redundancy. Many different control tasks can be formulated very elegantly in this formalism. Some examples (non-exhaustive): Position control, position and velocity control, grasp synthesis and control. Additionally Grupen et. al. researched discretization of the state of these controllers to provide a state space which is small enough to be feasible for reinforcement learning algorithms. This project aims at researching alternative state representations besides investigating the implications that concurrent execution pose for the controller learning task.