Barkour: Benchmarking Animal-level Agility with Quadruped Robots

TL;DR

Barkour introduces a dog-inspired agility benchmark for quadruped robots, combining a standardized obstacle course with a single intuitive score to quantify overall agility. The authors propose two baselines: specialist policies for individual tasks coordinated by a high-level navigator, and Locomotion-Transformer, a generalist policy distilled from specialists via a Transformer. Through extensive sim-to-real experiments on a custom hardware dog-like robot, they show that the generalist can achieve animal-level agility with smoother transitions, though it trails specialist performance and dogs on peak speed. The work highlights the value of Barkour as a research-friendly testbed and identifies concrete directions to close the remaining gap to canine performance, including onboard sensing, smarter hardware, and higher-level planning. Overall, Barkour provides a practical framework for benchmarking agility, generalization, and sim-to-real transfer in quadruped locomotion.

Abstract

Animals have evolved various agile locomotion strategies, such as sprinting, leaping, and jumping. There is a growing interest in developing legged robots that move like their biological counterparts and show various agile skills to navigate complex environments quickly. Despite the interest, the field lacks systematic benchmarks to measure the performance of control policies and hardware in agility. We introduce the Barkour benchmark, an obstacle course to quantify agility for legged robots. Inspired by dog agility competitions, it consists of diverse obstacles and a time based scoring mechanism. This encourages researchers to develop controllers that not only move fast, but do so in a controllable and versatile way. To set strong baselines, we present two methods for tackling the benchmark. In the first approach, we train specialist locomotion skills using on-policy reinforcement learning methods and combine them with a high-level navigation controller. In the second approach, we distill the specialist skills into a Transformer-based generalist locomotion policy, named Locomotion-Transformer, that can handle various terrains and adjust the robot's gait based on the perceived environment and robot states. Using a custom-built quadruped robot, we demonstrate that our method can complete the course at half the speed of a dog. We hope that our work represents a step towards creating controllers that enable robots to reach animal-level agility.

Figures (19)

• Figure 1: Barkour benchmark overview and example behavior.
• Figure 2: Barkour course design composed of four different obstacles: (start and end) pause tables, weave poles, an A-frame, and a broad jump.
• Figure 3: Overview of our methods to establish strong baselines for the Barkour benchmark. Top: Omni-directional walking, slope, and jumping policies are trained in simulation using RL. We then run the policies to create datasets which we use to distill a generalist Locomotion-Transformer policy. Bottom left: Switching between specialist policies using a hierarchical controller. Bottom right: generalist (Locomotion-Transformer) trained by distilling multiple specialist policies.
• Figure 4: An example set of way points used for the navigation controller. The robot stands on the start table and its heightfield is illustrated with blue rays hitting the floor.
• Figure 5: Custom small quadruped robot for hardware evaluation.