Agile perceptive multi-skill locomotion for quadrupedal robots in the wild

2026-07-15

AI Summary

This work addresses the challenge of achieving high-speed, multi-skilled, perception-driven agile locomotion and smooth gait transitions for quadrupedal robots in complex natural environments. The authors propose the APT-RL framework, which integrates an Action Pre-trained Transformer with reinforcement learning. By leveraging a large-scale 2D motion dataset generated through trajectory optimization, the framework pre-trains a transferable, high-quality motion prior. This prior is then combined with onboard perception and a simplified dynamics model to enable efficient policy learning and deployment on 3D rough terrain. Using only a single policy, the method robustly navigates diverse obstacles—including stairs, steps, and gaps—and achieves a peak speed of 6 m/s, significantly enhancing the robot’s agility and autonomy in both indoor and outdoor complex settings.

Abstract

Enabling quadrupedal robots to traverse complex terrains-from rugged outdoor environments to urban landscapes-requires seamless integration of multiple motor skills, smooth transitions between gaits, and high-speed perceptive locomotion using only onboard sensors. We present APT-RL (Action Pretrained Transformer-based Reinforcement Learning), a unified framework that enables multi-skill locomotion to achieve high-speed traversal in complex environments through autonomous skill transitions utilizing only onboard perception and computation. Our approach generates large-scale, feature-rich 2D motion datasets through trajectory optimization with simplified dynamics. These datasets enable training of diverse, reusable locomotion skills that transfer effectively to a real quadruped robot operating on complex uneven terrains. The resulting high-quality skills serve as strong priors for efficient learning of complex downstream tasks and extend naturally to 3D environments, enabling smooth, high-speed multi-skill locomotion in deployed policy. Real-world experiments demonstrate the framework's capabilities: the robot performs agile maneuvers through complex indoor obstacles and outdoor wild environments, including dynamic drop-down maneuvers that reach instantaneous peak speeds of up to 6 meters per second. A single onboard policy enables robust traversal of diverse obstacles, including stairs, hurdles, stepping stones, gaps, and fallen branches, demonstrating the versatility and effectiveness of our approach.

Problem

• quadrupedal robots
• multi-skill locomotion
• complex terrains
• onboard perception
• agile locomotion

Innovation

• multi-skill locomotion
• onboard perception
• trajectory optimization
• reinforcement learning
• quadrupedal robot