ANYmal parkour: Learning agile navigation for quadrupedal robots

By David Hoeller, N. Rudin, Dhionis V. Sako, Marco Hutter. Science Robotics, 2023

417 citations recorded by the source corpus.

Performing agile navigation with four-legged robots is a challenging task because of the highly dynamic motions, contacts with various parts of the robot, and the limited field of view of the perception sensors. Here, we propose a fully learned approach to training such robots and conquer scenarios that are reminiscent of parkour challenges. The method involves training advanced locomotion skills for several types of obstacles, such as walking, jumping, climbing, and crouching, and then using a high-level policy to select and control those skills across the terrain. Thanks to our hierarchical formulation, the navigation policy is aware of the capabilities of each skill, and it will adapt its behavior depending on the scenario at hand. In addition, a perception module was trained to reconstruct obstacles from highly occluded and noisy sensory data and endows the pipeline with scene understanding. Compared with previous attempts, our method can plan a path for challenging scenarios without expert demonstration, offline computation, a priori knowledge of the environment, or taking contacts explicitly into account. Although these modules were trained from simulated data only, our real-world experiments demonstrate successful transfer on hardware, where the robot navigated and crossed consecutive challenging obstacles with speeds of up to 2 meters per second. A learning-based approach performs agile navigation, perception, and locomotion with a quadrupedal robot at high speeds. Editor’s summary Agility in legged robots that match humans and animals is not easily achievable. Moreover, the ability to perform elegant and nimble locomotion around complex obstacles with limited onboard computing makes agility even more challenging. Hoeller et al. developed a framework for training a quadrupedal robot with locomotion skills, such as jumping, climbing, crouching, and walking, for rapid navigation around an obstacle parkour course. The framework was trained in simulation and subsequently deployed in the real world on legged robots, demonstrating their ability to reach targets with speeds of up to 2 meters per second and showing potential for robot navigation on unstructured terrain where time is vital, such as in search and rescue. —Amos Matsiko

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/31d83fd3f85d316012c7aa7c913a111065be0a3f

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (2024, background): “Recent research has explored DRL frameworks integrating locomotion with navigation, achieving high-speed obstacle avoidance (100) and agile navigation over challenging terrains (e.g., stairs, gaps, and boxes) (20, 96) and through confined 3D space (98, 99).”
Extreme Parkour with Legged Robots (2023, neutral): “[3] demonstrates agile behaviors by training task-specific policies and composing them using a high-level trained module but still relying on elevation maps.”