Learning Spatio-Temporal Planning from a Dynamic Programming Teacher: Feed-Forward Neurocontrol for Moving Obstacle Avoidance

Abstract

Within a simple test-bed, application of feed-forward neurocontrol for short-term planning of robot trajectories in a dynamic environ(cid:173) ment is studied. The action network is embedded in a sensory(cid:173) motoric system architecture that contains a separate world model. It is continuously fed with short-term predicted spatio-temporal obstacle trajectories, and receives robot state feedback. The ac(cid:173) tion net allows for external switching between alternative plan(cid:173) ning tasks. It generates goal-directed motor actions - subject to the robot's kinematic and dynamic constraints - such that colli(cid:173) sions with moving obstacles are avoided. Using supervised learn(cid:173) ing, we distribute examples of the optimal planner mapping over a structure-level adapted parsimonious higher order network. The training database is generated by a Dynamic Programming algo(cid:173) rithm. Extensive simulations reveal, that the local planner map(cid:173) ping is highly nonlinear, but can be effectively and sparsely repre(cid:173) sented by the chosen powerful net model. Excellent generalization occurs for unseen obstacle configurations. We also discuss the limi(cid:173) tations of feed-forward neurocontrol for growing planning horizons.