When Does Legacy Data Start to Help? Emergent Transfer in Cross-Configuration Robot Learning

2026-07-28

AI Summary

This study addresses the challenge of effectively transferring legacy demonstration data to a robot after hardware upgrades, specifically when only the camera and gripper are replaced while preserving the overall morphology across two generations of wheeled humanoid robots. Through real-world manipulation tasks, the authors identify a three-phase pattern in the utility of legacy data: it is initially ineffective at low policy capability, becomes significantly beneficial once a transfer threshold is surpassed, and then exhibits diminishing returns at high capability levels. The work presents the first theoretical explanation of this “transfer threshold” phenomenon and proposes a phase-aware data reuse strategy based on gradient alignment and policy uncertainty. The validity of the observed pattern and the efficacy of the proposed method are demonstrated on tasks including flower insertion (success rate improved from 23.3% to 86.7%), pen insertion (85.0% to 93.3%), and a mobile dual-arm watering task.

Abstract

Robotic hardware evolves over time, but demonstration data is often tied to a specific sensor and actuator configuration. This raises a practical and underexplored question: when does legacy data begin to benefit an upgraded robot? We study this question on a wheeled humanoid platform across two hardware generations, where both the camera and gripper are changed while the overall morphology remains fixed. Contrary to the common assumption that more cross-configuration data is always helpful, we observe a grokking-like transition: legacy data remains ineffective until the upgraded configuration acquires a minimum level of task competence, after which co-training gains rise sharply before diminishing near saturation. We hypothesize that this task-dependent transition is governed by a transfer threshold and characterize the resulting three-phase pattern. Across real-robot manipulation tasks, we observe all three phases: no measurable benefit at low competence ($10.0\% \rightarrow 10.0\%$), a sharp gain after crossing the threshold ($23.3\% \rightarrow 86.7\%$ on flower insertion), and diminishing returns at high competence ($85.0\% \rightarrow 93.3\%$ on pen insertion). We provide a theoretical account based on gradient alignment and residual policy uncertainty, and derive a phase-aware rule for deciding when to collect more new-hardware data and when to reuse legacy demonstrations. We further validate this three-phase pattern on a mobile dual-arm watering task, with results consistent with our predictions.

Problem

• legacy data
• cross-configuration transfer
• robot learning
• hardware evolution
• transfer threshold

Innovation

• cross-configuration transfer
• legacy data reuse
• transfer threshold
• grokking-like transition
• phase-aware learning