Quadrupedal and Legged Robot Locomotion

The field of robotics is witnessing significant advancements in quadrupedal and legged robot locomotion, with a focus on developing generalist approaches that can navigate complex and unstructured terrain. Reinforcement learning is emerging as a key solution, enabling optimal control through trial and error and allowing robots to learn from their environment. This approach is showing promise in overcoming the limitations of traditional legged robots, such as slippage and tripping, and is being applied to a range of scenarios, including parkour, climbing, and pick-and-place tasks. Notable papers in this area include:

  • Acrobotics, which presents a generalist reinforcement learning algorithm for quadrupedal agents in dynamic motion scenarios, achieving state-of-the-art results with reduced training agents.
  • Multi-Embodiment Locomotion at Scale, which demonstrates a single general locomotion policy trained on a diverse collection of 50 legged robots, achieving zero-shot transfer to unseen real-world humanoid and quadruped robots.

Sources

Acrobotics: A Generalist Approahc To Quadrupedal Robots' Parkour

Multi-Embodiment Locomotion at Scale with extreme Embodiment Randomization

CTBC: Contact-Triggered Blind Climbing for Wheeled Bipedal Robots with Instruction Learning and Reinforcement Learning

Learning Multi-Stage Pick-and-Place with a Legged Mobile Manipulator

Learning to Walk in Costume: Adversarial Motion Priors for Aesthetically Constrained Humanoids

Built with on top of