LHNav: Long-Horizon Humanoid Navigation via Multirate Coordination and Physics-Grounded Action Data
LHNav: Long-Horizon Humanoid Navigation via Multirate Coordination and Physics-Grounded Action Data
This paper introduces LHNav, a system designed for long-horizon humanoid navigation using multirate coordination and physics-grounded action data. The authors propose an architecture with three roles—Brain for task reasoning, Cerebellum for local motion generation, and Brainstem for whole-body execution—to maintain progress across successive goals while ensuring natural, stable motion. A physics-grounded data engine is developed to train the Cerebellum on collision-aware routes paired with rendered visual observations. In simulation, LHNav outperforms OmniNav and DualVLN in single-goal success and motion smoothness. Using a VLA trained only on single-goal demonstrations, LHNav achieves a 75.56% success rate on long-horizon navigation tasks. Deployment on the Unitree G1 demonstrates zero-shot long-horizon task execution in real-world environments with repositioned objects.
The authors present LHNav, an asynchronous system combining task reasoning, local motion generation, and whole-body execution to achieve long-horizon navigation for humanoids in simulation and on the Unitree G1 robot.
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
75.56%
Reported trials
View original evidence
t and extend single-goal navigation to successive destinations with natural, route-faithful whole-body motion. In simulation, our navigation VLA outperforms OmniNav and DualVLN in single-goal success and motion smoothness, with fewer collisions and falls. Using a VLA trained only on single-goal demonstrations, LHNav achieves a 75.56% success rate on long-horizon navigation tasks. Deployment on the Unitree G1 further
Open source S5
- paper: not_reported
- weights: not_reported
- dataset: not_reported
- code: 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.
t and extend single-goal navigation to successive destinations with natural, route-faithful whole-body motion. In simulation, our navigation VLA outperforms OmniNav and DualVLN in single-goal success and motion smoothness, with fewer collisions and falls. Using a VLA trained only on single-goal demonstrations, LHNav achieves a 75.56% success rate on long-horizon navigation tasks. Deployment on the Unitree G1 further
Open source S5
demonstrates single-goal navigation and zero-shot long-horizon task execution in real-world environments, including layouts with repositioned objects. Project website: https://robbyant-research.github.io/LHNav/.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org