A Compositional Neuro-Controller for Advanced Motor Control Tasks
This addresses the problem of efficient incremental learning and style switching in robotics, though it appears incremental as it builds on modular biological principles.
The authors tackled the challenge of building a robot controller that can incrementally learn and switch between motion styles by proposing a biologically inspired compositional neuro-controller, demonstrating that it reproduces trajectories more efficiently and learns different locomotion styles in a simulated robot-snake.
Humans and animals developed a sophisticated motor control apparatus and there is much evidence that it has a modular structure. The modularity offers a range of benefits, e.g. ability to learn dissociable motion styles without interference and forgetting, fast adaptation and de-adaptation to changes conditions. However in robotics, building a controller that can efficiently incrementally learn new motion styles and provide switching between them is a formidable challenge. In this paper we address the problem by proposing a novel biologically inspired compositional neuro-controller. We have shown that the compositional controller is able to reproduce a set of trajectories more efficiently comparing to a simple controller, exploiting incremental learning benefits. Second, we have demonstrated that the proposed controller is able to learn different locomotion styles and switch between them in a simulated robot-snake.