High-speed control and navigation for quadrupedal robots on complex and discrete terrain

High-speed legged navigation in discrete and geometrically complex
environments is a highly challenging task due to the
high-degree-of-freedom dynamics and long-horizon, non-convex nature of the
optimization problem. In this work, we propose a hierarchical navigation
pipeline for legged robots that is capable of traversing such environments
at high speed. The proposed pipeline consists of a planner and tracker
module. The planner module finds physically feasible foothold plans by
sampling-based optimization strategy, which involves sequential filtering.
This filtering process utilizes multiple criteria, including simple
heuristics and a learned neural network, to quickly eliminate bad samples.
Subsequently, rollouts are performed in a physics simulation to identify
the best foothold plan concerning the engineered cost function and to
confirm their physical consistency. This hierarchical planning module is
computationally efficient and physically accurate at the same time. The
tracker aims to accurately step on the target footholds from the planning
module. During the training stage, the foothold target distribution is
given by a generative model which is trained adversarially with the
tracker. This process ensures that the tracker is trained in a
sufficiently difficult environment. The resulting tracker is capable of
overcoming terrains that are more difficult than what the previous methods
could manage. We demonstrate this using Raibo, our in-house dynamic
quadrupedal robot. The results are highly dynamic and agile motions: Raibo
is capable of running on vertical walls, jumping a 1.3m gap, running over
stepping stones at 4 m/s, and autonomously navigating on terrains full of
30-degree ramps, stairs, and boxes of various sizes.