Probabilistic Reachable-Action Verification of Visuomotor Policies via Set-Based Training
2026-08-03 • Robotics
Robotics
AI summaryⓘ
The authors address the challenge of understanding all possible actions a robot can take from visual input, which is hard because big image processors slow things down. They keep the image processor fixed and analyze a simpler, smaller part of the system to make predictions more efficient. Using this approach, they improve how precisely they can predict action ranges while still completing tasks successfully. They also develop a way to give a probability-based boundary on these actions when the camera view changes. Their experiments confirm that their method is better at narrowing down possible actions than other comparison methods.
reachability analysisvisuomotor policiesvisual encoderset propagationzonotopescamera-pose perturbationsset-based trainingsplit conformal calibrationrobotic manipulation
Authors
Yanliang Huang, Zhuocheng Zhang, Peng Xie, Zhen Zhang, Wenyuan Wu, Majid Khadiv, Zhuoqi Zeng, Amr Alanwar
Abstract
Reachability analysis for visuomotor policies is difficult because large visual encoders make end-to-end set propagation computationally expensive and excessively conservative. We therefore freeze the visual encoder and confine set propagation to a low-dimensional interface between it and the downstream policy, with the interface set calibrated from held-out camera-pose perturbations. Propagating this set through the policy with zonotopes yields a terminal output-enclosure width that set-based training optimizes directly. During evaluation, camera-pose perturbations are sampled from the prescribed distribution, and rollout-level split conformal calibration converts the resulting action-deviation scores into a probabilistic reachable-action radius with finite-sample coverage. In controlled manipulation experiments, set-based training reduces this radius while preserving closed-loop task capability, and matched behavior-only, observational-consistency, and pointwise-adversarial controls all leave a larger radius.