ReSteer: Quantifying and Refining the Steerability of Multitask Robot Policies

TL;DR

ReSteer is a framework that quantifies and enhances the steerability of multitask robot policies, enabling robots to better respond to new instructions through a combination of metrics, data generation, and policy refinement.

Contribution

This work introduces ReSteer, a novel approach that measures and improves task steerability in multitask policies using a new proxy metric and self-refinement techniques.

Findings

ReSteer improves steerability by 11% in simulation.

Enhanced steerability enables robots to follow arbitrary instructions in real-world settings.

The framework reveals that steerability correlates with training data distribution overlap.

Abstract

Despite strong multi-task pretraining, existing policies often exhibit poor task steerability. For example, a robot may fail to respond to a new instruction ``put the bowl in the sink" when moving towards the oven, executing ``close the oven", even though it can complete both tasks when executed separately. We propose ReSteer, a framework to quantify and improve task steerability in multitask robot policies. We conduct an exhaustive evaluation of state-of-the-art policies, revealing a common lack of steerability. We find that steerability is associated with limited overlap among training task trajectory distributions, and introduce a proxy metric to measure this overlap from policy behavior. Building on this insight, ReSteer improves steerability via three components: (i) a steerability estimator that identifies low-steerability states without full-rollout evaluation, (ii) a steerable data generator that synthesizes motion segments from these states, and (iii) a self-refinement pipeline that improves policy steerability using the generated data. In simulation on LIBERO, ReSteer improves steerability by 11\% over 18k rollouts. In real-world experiments, we show that improved steerability is critical for interactive use, enabling users to instruct robots to perform any task at any time. We hope this work motivates further study on quantifying steerability and data collection strategies for large robot policies.