Collision-backpropagation based obstacle avoidance method for a legged robot expressed as a simplified dynamics model

Abstract

Simplified dynamics models have been widely used to make motion planning for legged robots less computationally complex. On the other hand, there hasn't been much study on collision avoidance for a simplified dynamics model. We present the collision-backpropagation based obstacle avoidance method (CBOA) to contribute to this problem. Our method uses the gradient flow of the collision cost to optimize the trajectory and prevent collisions with obstacles. According to our experiment, the CBOA can reduce planned trajectory collisions by up to 15.89 times compared to a previous implicit collision avoidance technique.