Lunar Rover's Behavior Fusion Learning Based on Nonholonomic Dynamics

摘要

The behavior-based motion planning with nonholonomic constrains for lunar rovers is discussed in this paper. For each fuzzy behavior controller, the hybrid coordinate scheme which combines competition with cooperation is proposed to ensure both the robustness and optimization. Force and torque of the wheels are chosen as the output of fuzzy behavior. The on-line Q-learning is used to obtain the behavior's coordinate scheme, ant its output is an optimal solution within a behavior decision set which is obtained according to the outputs from each behavior controller. Maggi equations are introduced to formulate the rover's transfer under the learnt controls. Experiment results demonstrate the effectiveness of this method and traceability of the trajectory.