Model Predictive Obstacle Avoidance Control Suppressing Expectation of Relative Velocity against Obstacles

In this paper, we propose a vehicle controller which attains practical obstacle avoidance while suppressing relative velocity against moving obstacles with uncertainty. For safety locomotion, obstacle avoidance is important but the observed position of obstacles are usually contaminated by noise while their motion inevitably possesses uncertainty. Especially in the crowded or narrow environment, it is sometimes impossible to find a path which completely prevents a vehicle from collision with surrounding obstacles. To resolve this issue, in this study, we propose a stochastic model predictive controller which alleviates the damage of potential collision by suppressing the expectation of relative velocity against obstacles while achieving avoidance with limited possibility of collision. In path planning, a vehicle tries to avoid the obstacles allowing limited possibility of collision using the covariance computed from the Kalman filter. Then, calculating a joint probability for the obstacle's position and velocity, the expectation of relative velocity between the vehicle and the obstacles is minimized. We verified the effectiveness of the proposed method through numerical simulations and an experiment of obstacle avoidance in a narrow corridor.