Versatile Locomotion Skills for Hexapod Robots

TL;DR

The paper tackles enabling a low-cost hexapod to autonomously navigate cluttered indoor environments by learning three skills—stair climbing, obstacle avoidance, and squeezing under obstacles—using only a depth camera and visual-inertial odometry. It introduces a two-stage teacher-student reinforcement learning framework where privileged information guides the teacher in simulation and is distilled into a depth/pose-based student for real hardware deployment, achieving zero-shot sim-to-real transfer. Through curriculum-based training in Isaac Gym and careful task-specific terrain design, the authors demonstrate robust performance across stairs, various obstacles, and squeeze scenarios on a $600$ SpiderPi platform with onboard perception, validated by real-world experiments. The work advances practical legged robotics by enabling complex locomotion in low-cost platforms without real-time joint feedback, with potential impact for home and attic navigation tasks.

Abstract

Hexapod robots are potentially suitable for carrying out tasks in cluttered environments since they are stable, compact, and light weight. They also have multi-joint legs and variable height bodies that make them good candidates for tasks such as stairs climbing and squeezing under objects in a typical home environment or an attic. Expanding on our previous work on joist climbing in attics, we train a legged hexapod equipped with a depth camera and visual inertial odometry (VIO) to perform three tasks: climbing stairs, avoiding obstacles, and squeezing under obstacles such as a table. Our policies are trained with simulation data only and can be deployed on lowcost hardware not requiring real-time joint state feedback. We train our model in a teacher-student model with 2 phases: In phase 1, we use reinforcement learning with access to privileged information such as height maps and joint feedback. In phase 2, we use supervised learning to distill the model into one with access to only onboard observations, consisting of egocentric depth images and robot pose captured by a tracking VIO camera. By manipulating available privileged information, constructing simulation terrains, and refining reward functions during phase 1 training, we are able to train the robots with skills that are robust in non-ideal physical environments. We demonstrate successful sim-to-real transfer and achieve high success rates across all three tasks in physical experiments.

Figures (10)

• Figure 1: Physical and URDF of the robot. (a) The hexapod robot standing at the reset position - roughly 37 cm tall. (b) The URDF of the hexapod.
• Figure 2: High-level overview of training methodology 10341957
• Figure 3: Simulation environment in Isaac Gym. (a) Stairs Climbing. (b) Obstacle Avoidance. (c) Squeezing under obstacles.
• Figure 4: Reward vs. episode convergence curve for tasks. (a) Stairs Climbing. (b) Obstacle avoidance. (c) Squeezing under obstacles.
• Figure 5: Physical design of a squeezing environment.