FlowCorrect: Efficient Interactive Correction of Generative Flow Policies for Robotic Manipulation

2026-02-25

AI Summary

This work addresses the vulnerability of generative manipulation policies to distribution shifts during deployment, which often leads to catastrophic failures—many of which are “near-miss” cases amenable to minor corrections. To remedy this without retraining the backbone policy, the authors propose FlowCorrect, a framework that enables sparse, online policy refinement through minimal human-provided pose nudges. Leveraging a lightweight VR interface, the method collects extremely limited human interaction data and dynamically adjusts action outputs via a local adaptation mechanism. Evaluated on three real-world tabletop tasks—pick-and-place, pouring, and uprighting a tipped cup—FlowCorrect achieves an 85% improvement in success rate under challenging conditions with a minimal correction budget, while preserving performance in originally successful scenarios. This significantly enhances deployment robustness and sample efficiency.

Technology Category

• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Generative manipulation policies can fail catastrophically under deployment-time distribution shift, yet many failures are near-misses: the robot reaches almost-correct poses and would succeed with a small corrective motion. We present FlowCorrect, a deployment-time correction framework that converts near-miss failures into successes using sparse human nudges, without full policy retraining. During execution, a human provides brief corrective pose nudges via a lightweight VR interface. FlowCorrect uses these sparse corrections to locally adapt the policy, improving actions without retraining the backbone while preserving the model performance on previously learned scenarios. We evaluate on a real-world robot across three tabletop tasks: pick-and-place, pouring, and cup uprighting. With a low correction budget, FlowCorrect improves success on hard cases by 85\% while preserving performance on previously solved scenarios. The results demonstrate clearly that FlowCorrect learns only with very few demonstrations and enables fast and sample-efficient incremental, human-in-the-loop corrections of generative visuomotor policies at deployment time in real-world robotics.

Problem

• generative manipulation policies
• distribution shift
• near-miss failures
• human-in-the-loop correction
• deployment-time adaptation

Innovation

• interactive correction
• generative flow policies
• human-in-the-loop
• deployment-time adaptation
• sample-efficient learning