Try Once, Then Optimal: De-Redundified Procedure Memory for Cross-Episode Exploration Amortization

2026-07-26

AI Summary

This work addresses the inefficiency of robotic manipulation when interacting with objects that possess hidden internal states, where repeated exploration of identical instances leads to redundant effort. The authors propose an Instance-Oriented Memory (IOM) framework that, for the first time, anchors cross-episode memory to object instances rather than task states. By extracting manipulation procedures from a single interaction and storing them using visual features of the object as keys, the system enables direct reuse upon encountering the same instance, thereby amortizing exploration costs. The approach leverages off-the-shelf vision-language models (VLMs) to automatically parse manipulation sequences and integrates programmatic conditional policies to inject soft priors. Experiments in both simulated and real-world settings demonstrate a 16–30% reduction in required manipulation steps, with the VLM-based variant recovering 69–88% of the efficiency gain while maintaining or even improving task success rates.

Abstract

Manipulating objects with hidden internal state, such as a latched microwave, forces a robot to probe before it can act. Yet a robot that has solved an instance once re-runs the same probes whenever it encounters that instance again, because existing cross-episode memories target task success and organize reuse around states, not the object or the cost of re-exploring it. We present Instance-Oriented Memory (IOM), an object-centric framework that amortizes this exploration: from a single encounter that uncovers the hidden state, whether or not it succeeds, IOM records a short procedure for manipulating that instance, keys it on the object's identifiable features, and injects it as a soft bias on a procedure-conditioned policy. A later encounter recognizes the object and recalls its procedure instead of re-exploring. We instantiate this distillation with an off-the-shelf vision-language model (VLM) that parses each encounter into the procedure without task-specific training. Across four articulated-object tasks, two in simulation (microwave, door) and two on a real robot (bottle, cabinet), an oracle procedure memory cuts manipulation operations by 16-30% over re-exploration at non-regressing success, and the VLM instantiation recovers 69-88% of that saving out of the box. Because the procedure is a soft bias on a feedback-driven policy, an incorrect memory is recovered from rather than obeyed: success holds even when a retrieved procedure is wrong, as for $\approx$12% of door instances. Across all tasks the benefit is purely one of efficiency: success never regresses, and on the real robot even improves. Code will be released upon acceptance.

Problem

• cross-episode memory
• hidden internal state
• object-centric manipulation
• exploration amortization
• procedure reuse

Innovation

• Instance-Oriented Memory
• cross-episode exploration amortization
• procedure memory
• vision-language model
• object-centric learning