Locomotion

A robot's ability to move itself from place to place, using legs, wheels, or other means.

Locomotion broadly covers the ways a robot moves itself: rolling on wheels, walking on legs, jumping, flying. Together with manipulation, which changes the state of external objects, it forms one of the two main branches of robot capability. In embodied AI, the term usually means legged locomotion in quadruped, biped, and humanoid robots: coordinating a dozen or more joints, keeping balance, and handling terrain like stairs and loose gravel. Traditional methods relied on dynamics models and model predictive control; the dominant recent approach trains a policy with reinforcement learning inside a GPU-parallel simulator, then transfers it to the real robot. A 2021 project from ETH Zurich and NVIDIA simulated thousands of ANYmal quadrupeds at once on a single GPU, training a flat-ground walking policy in under four minutes and a rough-terrain policy in about 20 minutes. In Chinese, 运动 also commonly corresponds to “motion” in general; “locomotion” specifically means moving the body itself.

ExampleUsing legged_gym inside Isaac Gym to simulate 4,096 ANYmal quadrupeds at once, training for about 20 minutes on terrain like stairs, slopes, and obstacles produces a policy that deploys straight to the real robot — able to climb stairs and cross obstacles.

Related

Legged Locomotion · Bipedal Locomotion · RL-based Locomotion Control · Sim-to-Real Transfer · Loco-manipulation · Manipulation

Sources

Robot locomotion (Wikipedia)
Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement Learning (arXiv:2109.11978)

As of

2021-09