STRIDE: Learning Humanoid Locomotion in Minutes with Acceleration-Based Rewards
STRIDE: Learning Humanoid Locomotion in Minutes with Acceleration-Based Rewards
This paper introduces STRIDE, a method for learning humanoid locomotion directly on physical robots by using acceleration-based rewards instead of base velocity tracking. By combining off-policy reinforcement learning (XQC), high-gain joint controllers, and symmetry augmentation, STRIDE learns to walk on Unitree G1 and Booster T1 humanoids in simulation within 14 minutes of samples. The approach reduces training falls by 81% compared to methods without these components, which typically require over an hour to learn.
The authors propose STRIDE, a reinforcement learning method that uses acceleration-based rewards and off-policy algorithms to learn humanoid locomotion in simulation within minutes using only onboard IMU data, reporting an 81% reduction in training falls compared to baseline approaches.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S3
Reported numbers
Reported duration
14 minutes
View original evidence
TRIDE learns to walk on the Unitree G1 and Booster T1 humanoids within 14 minutes of samples. Compared to learning without these two components, it reduces the falls during training by 81%, while without them the robot learns only in over an hour.
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
otion from minutes of samples with a reward that only requires the onboard Inertial Measurement Unit (IMU). Instead of tracking the base velocity, STRIDE rewards the robot for following a commanded acceleration profile. To learn efficiently from this reward, we use the off-policy RL algorithm XQC and combine a high gain setting for the joint position controllers with left-right symmetry augmentation. In simulation, S
Open source S3
TRIDE learns to walk on the Unitree G1 and Booster T1 humanoids within 14 minutes of samples. Compared to learning without these two components, it reduces the falls during training by 81%, while without them the robot learns only in over an hour.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org