Horizon-Coupled Obstacle Guidance for Relative-Action Diffusion Policies
Horizon-Coupled Obstacle Guidance for Relative-Action Diffusion Policies
This research introduces an inference-time obstacle-guidance framework for pretrained relative-action diffusion policies that propagates obstacle responses across the action horizon to coordinate task-directed avoidance. Experiments on real robots across seven obstacle-constrained scenarios show the method achieves 105 out of 140 (75.0%) collision-free task successes, compared to 8 out of 140 (5.7%) for absolute-action pointwise guidance. Ablation studies indicate that the relative-action horizon-coupled formulation yields larger improvements across both tested obstacle models.
The study presents a new inference-time obstacle-guidance framework for pretrained relative-action diffusion policies, demonstrating improved collision-free success rates in real-robot experiments across seven scenarios; however, the results are specific to the tested environments and do not constitute general deployment readiness.
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
total trials
140
View original evidence
pagates each obstacle response to all contributing action increments during denoising. Real-robot experiments across seven obstacleconstrained scenarios spanning four manipulation tasks show that our method achieves 105/140 (75.0%) collision-free task successes, compared with 8/140 (5.7%) for absolute-action pointwise guidance. Ablations show that the relative-action horizoncoupled formulation provides the larger obs
Open source S4scenarios
7
View original evidence
pagates each obstacle response to all contributing action increments during denoising. Real-robot experiments across seven obstacleconstrained scenarios spanning four manipulation tasks show that our method achieves 105/140 (75.0%) collision-free task successes, compared with 8/140 (5.7%) for absolute-action pointwise guidance. Ablations show that the relative-action horizoncoupled formulation provides the larger obs
Open source S4manipulation tasks
4
View original evidence
pagates each obstacle response to all contributing action increments during denoising. Real-robot experiments across seven obstacleconstrained scenarios spanning four manipulation tasks show that our method achieves 105/140 (75.0%) collision-free task successes, compared with 8/140 (5.7%) for absolute-action pointwise guidance. Ablations show that the relative-action horizoncoupled formulation provides the larger obs
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
pagates each obstacle response to all contributing action increments during denoising. Real-robot experiments across seven obstacleconstrained scenarios spanning four manipulation tasks show that our method achieves 105/140 (75.0%) collision-free task successes, compared with 8/140 (5.7%) for absolute-action pointwise guidance. Ablations show that the relative-action horizoncoupled formulation provides the larger obs
Open source S4
arXiv:2611.09340v1 Announce Type: new Abstract: Real-world robot deployment requires adapting learned policies to unseen obstacles without collecting new demonstrations or retraining. Existing policy-level guidance often applies obstacle corrections independently at individual horizon steps, which may address local obstacle violations but fail to coordinate the action sequence into a complete, task-directed detour. H
Open source S2
erved improvement across both obstacle models.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org