World-Model-Augmented Visual Locomotion for Humanoids on Foothold-Constrained Terrain
World-Model-Augmented Visual Locomotion for Humanoids on Foothold-Constrained Terrain
This research introduces WM-LOCO, a method that combines a recurrent state-space model (RSSM) with proximal policy optimization (PPO) to enable humanoids to navigate sparse and discontinuous terrains like gaps, stepping stones, stairs, and balance beams. The system uses onboard depth and proprioception to predict future states and rewards, improving foot placement accuracy. It achieves average success rates of 96.0% on gaps, 96.9% on stepping stones, 94.0% on stairs, and 92.1% on balance beams in simulation, with a zero-shot sim-to-real transfer to a Unitree G1 robot achieving 92.5% success across the same terrain classes.
The study demonstrates a world-model-augmented visual locomotion policy achieving high success rates on constrained terrains with zero-shot transfer to a Unitree G1 humanoid.
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
Reported success rate
96%
Reported zero-shot result; the excerpt does not specify an evaluation protocol.
Reported trials
View original evidence
ximal policy optimization (PPO) actor-critic, providing predictive context for anticipatory foot placement on foothold-constrained terrain. WM-LOCO achieves average success rates of 96.0% on gaps, 96.9% on stepping stones, 94.0% on stairs, and 92.1% on balance beams, outperforming the evaluated baselines while improving gait efficiency and smoothness. Furthermore, WM-LOCO achieves zero-shot sim-to-real transfer to a
Open source S4
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ximal policy optimization (PPO) actor-critic, providing predictive context for anticipatory foot placement on foothold-constrained terrain. WM-LOCO achieves average success rates of 96.0% on gaps, 96.9% on stepping stones, 94.0% on stairs, and 92.1% on balance beams, outperforming the evaluated baselines while improving gait efficiency and smoothness. Furthermore, WM-LOCO achieves zero-shot sim-to-real transfer to a
Open source S4
Unitree G1, with 92.5% success across these four terrain classes.
Open source S5
World-Model-Augmented Visual Locomotion for Humanoids on Foothold-Constrained Terrain
Open source S1
arXiv:2609.02542v2 Announce Type: replace Abstract: Foothold-constrained terrain limits feasible footholds to sparse, discontinuous, or narrow support regions, as in gaps, stepping stones, stairs, and balance beams. In such settings, a single misplaced step can cause a fall. Reliable traversal therefore requires precise foot placement and anticipation of upcoming terrain. To address this challenge, we present World-M
Open source S2
odel-Augmented Visual Locomotion (WM-LOCO), which augments a visual locomotion policy with predictive context from a jointly trained world model. Given onboard depth observations and proprioception, a recurrent state-space model (RSSM) learns action-conditioned latent dynamics that predict future proprioceptive states, egocentric observations, and task rewards. The resulting recurrent feature is then fed into the pro
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org