Deliberative planner for UGV with actively articulated suspension to negotiate geometric obstacles by using centipede locomotion pattern

In this paper, we propose deliberative upper-level behavior planning method for UGV with actively articulated suspension to negotiate geometric obstacle. Proposed deliberative planning method used Q-learning with the expert model for negotiating specific obstacle type. We modify the centipede locomotion pattern to the suspensions´ locomotion and define the MDP using it. In 2D space, we define the state, action and reward model, and apply the conservative ε-greedy action selection method to shorten the behavior plans that is transferred to lower level behavior planner. Also, we use the obstacle negotiation expert model to Q-learning because the state space is too large to solve by Q-learning. We show that Q-learning with expert model can improve the convergence properties in very large space of state and action, and our algorithm can generate the behavior plans for various dimensions of the step up obstacle by simulation environment in our goal time of 10 seconds.