Asymmetric physics enables efficient learning in quadrupedal robot swarms

2026-06-22

AI Summary

Existing approaches struggle to enable robot swarms to efficiently learn animal-like local coordination and navigation capabilities in complex physical environments. This work proposes an asymmetric physical training framework that integrates a high-fidelity, non-differentiable simulator with a differentiable surrogate model to achieve large-scale, decentralized, end-to-end control of quadrupedal robots using only monocular visual input. The method successfully trains navigation policies for 512 robots in obstacle-dense environments—the largest such demonstration to date—and achieves zero-shot transfer to six physical robots. The learned policies generalize effectively across diverse real-world scenarios, including forests, bridges, and mazes, exhibiting emergent collective behaviors such as predictive obstacle avoidance and right-of-way adherence in five distinct physical environments.

Abstract

Animal collectives navigate cluttered environments through local coordination, yet robot swarms still struggle to reproduce this capability in the physical world. End-to-end learning offers a route to such coordination, but scaling it to embodied swarms remains difficult: standard sampling-based reinforcement learning becomes inefficient when visual perception, dense robot-robot interaction, and contact-rich locomotion must be learned together. Here we show that asymmetric physics enables efficient end-to-end learning of vision-based, decentralized control in large swarms of quadrupedal robots. During training, quadrupeds interact in shared environments, where a high-fidelity, non-differentiable simulator generates realistic motion and contact dynamics, and differentiable surrogate models provide gradients for navigation and locomotion policies. This separation enables up to 512 quadrupeds to learn coordinated navigation policies in obstacle-rich environments. At deployment, each robot acts from a single forward-facing depth camera, without explicit communication, centralized planning, or global maps. The policies generalize across forests, bridges, enclosures, narrow passages, and mazes, and zero-shot transfer to six physical quadrupeds across five real-world scenarios. The resulting swarms exhibit predictive avoidance, right-side yielding, pausing before bottlenecks, and wall following, showing that asymmetric physics enables efficient training of scalable decentralized control policies for quadrupedal robot swarms.

Problem

• quadrupedal robot swarms
• decentralized control
• coordinated navigation
• obstacle-rich environments
• end-to-end learning

Innovation

• asymmetric physics
• quadrupedal robot swarms
• end-to-end learning
• differentiable surrogate models
• decentralized control