Modeling motor-perceptual behaviors to enable intuitive paths in an aquatic robot

This ongoing research focuses on providing skills based on motor-perceptual behaviors to an underwater vehicle in order to execute collision-avoidance trajectories in a more natural and intuitive way. An intuitive action can be seen as a reflex in humans and some animals. Reflexs do not involve a conscious reasoning at the time of execution, this is because a motorperceptual skill for a particular action has been already developed in the brain from a past execution. For the case of a robot, the analogy would be that, during a collision-avoidance action, no sensor feedback is needed to correct the robot's state, and thus there is no need of applying a control law to achieve the required locomotion. The direct consequence of this would be the reduction of robot cost while maintaining its performance. Our approach involves a training phase, in which a set of primitive curve trajectories, at different radius and velocities, parameterized in time and orientation, are performed by using a PD control. By applying a linear regression classifier, only those output control parameters that yield stable and accurate trajectories are fed to a knowledge database. These parameters are used to perform intuitive trajectories without using a control law. Preliminary results show the feasibility of our method.