Sleep & Wellness Guide
Do Not Cut When Uncertain: Rejectable and Calibrated Decision Heads for VLA Policies in Robotic Harvesting
Key Takeaway
A robotics research paper on Do Not Cut When Uncertain: Rejectable and Calibrated Decision Heads for VLA Policies in Robotic Harvesting.
Practical Tips
Practical tips and how-to guidance will be added by our editorial team.
中文解读
中文解读待补充:本站将优先为睡眠改善、失眠治疗、助眠方法等高价值文章补充中文说明。
Article Summary
Vision-Language-Action (VLA) policies trained with behavior cloning or flow matching are optimized to output an action trajectory, but they cannot express "I don't know" or "I should not act." In robotic harvesting, occlusion makes single-frame decisions fundamentally ambiguous: identical pixels can correspond either to a cuttable stem or to no stem at all. Existing VLAs are forced to commit, leading to high-confidence errors with irreversible consequences. We argue that the failure mode of a VLA is determined not by backbone scale but by its output interface. We propose Rejectable and Calibrated Decision Heads (RCDH), a typed, rejectable, and calibrated output interface that can be attached to a frozen VLA backbone without retraining or new features. RCDH introduces (i) a decision schema with explicit rejection and ordered, conditional decomposition, and (ii) a calibration procedure for risk-aware abstention. We evaluate RCDH on a robotic harvesting platform with controllable leaf occlusion, comparing generative, enumerated, calibrated, and rejectable interfaces. We show that replacing only the output head restores out-of-distribution usability under occlusion while preserving in-distribution performance. We further test whether the ordering of the rejection space is critical. Our results suggest that the right to refuse, rather than a larger model, is the missing interface for reliable manipulation under uncertainty.
Sources & References
Need to track a shipment?
Use our free logistics tracking tool to check real-time delivery status for USPS, FedEx, UPS, DHL, Amazon and 1000+ carriers worldwide.
Track a Package Now
Comments