aiqbench analysis
This research introduces INSPO, a method that optimizes input noise trajectories to steer generative robot policies in real-time while satisfying safety constraints like collision avoidance. By avoiding computationally expensive iterative diffusion processes, the approach achieves lower runtime costs and improved constraint satisfaction compared to existing gradient-guided or projection-based techniques. The study demonstrates significant gains in task success across diverse robotic benchmarks using both state-based and vision-language-action models.
Safe Real-Time Policy Steering via Noise-Space Trajectory Optimization for One-Step Generative Policies
arXiv:2609.21220v1 Announce Type: new Abstract: Generative robot policies can represent diverse, multimodal behaviors, but adapting pretrained policies to deployment-time constraints such as collision avoidance and orientation maintenance remains challenging. Existing inference-time steering methods typically apply gradient guidance through iterative diffusion or flow processes, which can be computationally expensive for real-time control. We propose INSPO, which formulates inference-time steering of one-step generative policies as trajectory optimization in the policy's input noise space. By optimizing the input noise while evaluating constraints on the induced state trajectory, INSPO searches the policy-induced behavior space without directly modifying generated actions. The optimization includes a regularization term that encourages solutions to remain consistent with the policy's input distribution and is solved online using population-based particle optimization. We evaluate INSPO on state- and image-based task-specific policies and generalist vision-language-action policies across Push-T, Can pick-and-place, and LIBERO-Spatial. INSPO improves task success and constraint satisfaction over best-of-N sampling and action projection, while comparing favorably with gradient-guided generation at lower runtime.
Abstract:Generative robot policies can represent diverse, multimodal behaviors, but adapting pretrained policies to deployment-time constraints such as collision avoidance and orientation maintenance remains challenging. Existing inference-time steering methods typically apply gradient guidance through iterative diffusion or flow processes, which can be computationally expensive for real-time control. We propose INSPO, which formulates inference-time steering of one-step generative policies as trajectory optimization in the policy's input noise space. By optimizing the input noise while evaluating constraints on the induced state trajectory, INSPO searches the policy-induced behavior space without directly modifying generated actions. The optimization includes a regularization term that encourages solutions to remain consistent with the policy's input distribution and is solved online using population-based particle optimization. We evaluate INSPO on state- and image-based task-specific policies and generalist vision-language-action policies across Push-T, Can pick-and-place, and LIBERO-Spatial. INSPO improves task success and constraint satisfaction over best-of-N sampling and action projection, while comparing favorably with gradient-guided generation at lower runtime.
Robotics (cs.RO)
arXiv:2609.21220 [cs.RO]
(or arXiv:2609.21220v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.21220 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 249 words
Related coverage
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · Aug 3, 2026
The Entangled Dimensions of Decision Theory
LessWrong · Jul 30, 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
Inside the suddenly explosive world of AI safety
The Verge · Sep 17, 2026