Curriculum choice affects quadruped robot learning success across terrains
Is It the Curriculum or Just the Easy Start? Benchmarking Unsupervised Environment Design on Quadrupeds
Machine LearningRobotics
Summary
Teaching four-legged robots to move over different surfaces depends a lot on how the training environments are designed. The authors compared many methods that create these environments automatically and by hand across five robot types. They found that no single training method works best for all robots or all terrains. Sometimes simpler or easier training tasks improve learning more than more complex ones. Their work helps decide which training approach to pick for a given robot and terrain type.
What this means in practice
- •For robotics engineers: Select the most effective environment curriculum for training specific quadruped robots to improve terrain navigation performance.
- •For simulation platform developers: Incorporate the benchmarked curriculum strategies to provide better default training scenarios for legged robot simulations.
Authors
Jakub Gilewicz, Dominik Galus, Piotr Wojdas
Abstract
Unsupervised environment design (UED) methods such as PLR, ACCEL, SFL and PAIRED are compared mostly on 2D mazes, grid worlds and BipedalWalker, while terrain curricula for legged robots are hand-designed and evaluated against a few baselines. We present a benchmark that runs fourteen UED, domain-randomisation and robotics curricula under one protocol on five 3D quadrupeds in MuJoCo MJX: the closed-chain W01-TEK and the open-chain Unitree Go1 and Go2, Google Barkour vB and Boston Dynamics Spot. Levels are vectors of eight interpretable terrain parameters, and all curricula drive the same PPO student at matched environment-step budgets. The evaluation reports success on random and held-out structured terrains, interquartile means with stratified bootstrap intervals, capability envelopes and a level-by-level domination comparison. We compare methods within each robot and report, per level set, the top group of methods not significantly below the best. No method is in the top group on every robot and level set. Methods with easy or low-difficulty training levels, curated or not, fill the top groups on in-distribution random levels of Go1 and Barkour, and of W01-TEK apart from PAIRED+BiBC. The success-driven terrain curriculum from legged-robot practice is in a top group most often, domain randomisation on uniform levels in none, and PAIRED in none outside Spot, where the fourteen methods lie within 3.8 points on random levels of difficulty 0.25. On three robots the leader on in-distribution random levels is not the leader on held-out terrain. We collect the results into a per-robot guide to choosing a curriculum.