aiqbench analysis
This research introduces REDACT, a teacher-student framework designed to maintain robust depth perception for autonomous locomotion when facing unmodeled sensor corruptions like occlusion or noise. By utilizing approximate conformal calibration on clean data, the system achieves zero-shot transfer to real-world environments without needing prior knowledge of specific corruption types. The work advances the reliability of embodied AI agents in unpredictable physical settings where standard synthetic augmentation fails.
REDACT: Robust Perceptive Locomotion under Unseen Visual Corruption
arXiv:2609.25450v1 Announce Type: new Abstract: Depth-conditioned locomotion policies have demonstrated impressive agile maneuvers, but can be steered to unpredictable actions when observations are outside their training distribution. Occlusion, invalid returns, sensor noise, and visual distractors can shift deployment observations away from nominal simulated depth. While synthetic sensor augmentation targets specified degradations, it does not by itself define behavior under corruption families omitted from training. To address gaps in training-time coverage, we present REDACT (Retaining Evidence Despite Artifacts for Continued Traversal), a teacher-student framework combining an improved visual encoder architecture, persistent feature masking, and a novel consensus-gating algorithm to retain useful depth information under unmodeled corruption. The gate uses approximate conformal calibration on clean observations alone, requiring no prior knowledge of the corruption type. Trained on clean simulated depth, REDACT retains useful visual information under unseen corruption, supporting higher traversal success than existing parkour baselines. Evaluation of depth augmentation across corruption families further shows that REDACT improves robustness where augmentation coverage is missing. Real-world trials demonstrate zero-shot transfer to structured and forested environments with unfamiliar scene content.
Abstract:Depth-conditioned locomotion policies have demonstrated impressive agile maneuvers, but can be steered to unpredictable actions when observations are outside their training distribution. Occlusion, invalid returns, sensor noise, and visual distractors can shift deployment observations away from nominal simulated depth. While synthetic sensor augmentation targets specified degradations, it does not by itself define behavior under corruption families omitted from training. To address gaps in training-time coverage, we present REDACT (Retaining Evidence Despite Artifacts for Continued Traversal), a teacher-student framework combining an improved visual encoder architecture, persistent feature masking, and a novel consensus-gating algorithm to retain useful depth information under unmodeled corruption. The gate uses approximate conformal calibration on clean observations alone, requiring no prior knowledge of the corruption type. Trained on clean simulated depth, REDACT retains useful visual information under unseen corruption, supporting higher traversal success than existing parkour baselines. Evaluation of depth augmentation across corruption families further shows that REDACT improves robustness where augmentation coverage is missing. Real-world trials demonstrate zero-shot transfer to structured and forested environments with unfamiliar scene content.
8 pages, 9 figures. Project page: this https URL
Robotics (cs.RO)
arXiv:2609.25450 [cs.RO]
(or arXiv:2609.25450v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.25450 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 · 273 words
Related coverage
GOT-JEPA: Generic Object Tracking with Model Adaptation and Occlusion Handling using Joint-Embedding Predictive Architecture
arXiv · Jun 17, 2026
Agile and Generalized Legged Locomotion via Attention-Based Neural Map Encoding
arXiv · Sep 9, 2026
Towards Dependable Optical Flow: Benchmarks for Real-World Spatiotemporal Corruptions
arXiv · Oct 14, 2026
Towards robust multimodal 3D object detection via visual foundation models
arXiv · Sep 22, 2026