Learning to Navigate by Pushing

In this work, we investigate a form of dynamic contact-rich locomotion in which a robot pushes off from obstacles in order to move through its environment. We present a reflex-based approach that switches between optimized hand-crafted reflex controllers and produces smooth and predictable motions. In contrast to previous work, our approach does not rely on periodic movements, complex models of robot and contact dynamics, or extensive hand tuning. We demonstrate the effectiveness of our approach and evaluate its performance compared to a standard model-free RL algorithm. We identify continuous clusters of similar behaviours, which allows us to successfully transfer different push-off motions directly from simulation to a physical robot without further retraining.

I. Introduction

Humans exploit highly dynamic interactions with the environment to reach states that would have otherwise been inaccessible. For instance, to move to an out-of-reach hand-hold, a rock climber jumps to the new handhold by pushing against the wall with their legs. Swinging from the current handhold before the jump can help them gain momentum for larger jumping distances. In parkour, a popular technique for scaling high walls is to run towards the wall and then jump onto and push off the wall with a foot to reach the top. In contrast to these highly dynamic interactions, robotic systems typically attempt to avoid obstacles and only navigate and operate in clear, structured settings such as labs and empty hallways. If robots can use their arms to dynamically push off obstacles, they could move more freely, flexibly, and rapidly in cluttered environments. However, this particular skill is difficult to learn because of the intermittent contact events we encounter during a dynamic pushing motion.