STEP: Spatiotemporal Training with Explicit Prediction for Perceptive Humanoid Locomotion
STEP: Spatiotemporal Training with Explicit Prediction for Perceptive Humanoid Locomotion
This research presents STEP, a framework for perceptive humanoid locomotion that jointly supervises cross-modal relationships and action-dependent state evolution. The method uses bidirectional masked prediction to learn terrain-body relationships and an action-conditioned predictor to map policy states. Evaluated in simulation and deployed on a 23-DoF Noetix E1 humanoid, STEP achieves successful traversal of stairs, platforms, gaps, and stepping stones, succeeding in 98 out of 100 hardware trials.
The paper introduces a one-stage framework that jointly supervises cross-modal relationships and action-dependent state evolution, demonstrating improved terrain traversal in simulation and achieving 98 out of 100 successful hardware trials on a 23-DoF Noetix E1 humanoid across specific obstacles including stairs and gaps.
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.
Reported numbers
robot_dof
23
View original evidence
mparisons show overall improvements in terrain traversal over evaluated fusion alternatives and ablations, while depth-dropout and delay experiments demonstrate the benefits of both predictive objectives under degraded visual input. Deployed on a 23-DoF Noetix E1 humanoid with onboard sensing, STEP succeeds in 98 of 100 hardware trials across stairs, a 40\,cm platform, a 70\,cm gap, stepping stones, and backward stai
Open source S5platform_height
40
View original evidence
mparisons show overall improvements in terrain traversal over evaluated fusion alternatives and ablations, while depth-dropout and delay experiments demonstrate the benefits of both predictive objectives under degraded visual input. Deployed on a 23-DoF Noetix E1 humanoid with onboard sensing, STEP succeeds in 98 of 100 hardware trials across stairs, a 40\,cm platform, a 70\,cm gap, stepping stones, and backward stai
Open source S5gap_width
70
View original evidence
mparisons show overall improvements in terrain traversal over evaluated fusion alternatives and ablations, while depth-dropout and delay experiments demonstrate the benefits of both predictive objectives under degraded visual input. Deployed on a 23-DoF Noetix E1 humanoid with onboard sensing, STEP succeeds in 98 of 100 hardware trials across stairs, a 40\,cm platform, a 70\,cm gap, stepping stones, and backward stai
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
mparisons show overall improvements in terrain traversal over evaluated fusion alternatives and ablations, while depth-dropout and delay experiments demonstrate the benefits of both predictive objectives under degraded visual input. Deployed on a 23-DoF Noetix E1 humanoid with onboard sensing, STEP succeeds in 98 of 100 hardware trials across stairs, a 40\,cm platform, a 70\,cm gap, stepping stones, and backward stai
Open source S5
STEP: Spatiotemporal Training with Explicit Prediction for Perceptive Humanoid Locomotion
Open source S1
Source：arXiv Robotics — research abstracts · arxiv.org