Hidden-state steering improves imitation policy success under noise
Diffeomorphic Hidden-State Steering: Improving Imitation Learning Policy Robustness through Expert Hidden-State Geometry
Robotics
Summary
Imitation learning policies can fail when small errors add up during use, especially if observations are noisy. The authors found that expert behaviors vary mostly along a few hidden directions that change with context. They created a method called Diffeomorphic Hidden-State Steering that adjusts a policy's hidden state during testing, based on this geometry, without retraining or new data. This approach helps the policy better handle noisy inputs and improves success rates in simulated robot tasks.
What this means in practice
- •For robotics engineers: Improve the reliability of robot manipulation policies in noisy real-world settings without extra training data.
- •For autonomous system developers: Enhance control robustness of learned controllers in systems where retraining policies frequently is infeasible.
Authors
KwangBin Lee
Abstract
Small observation or action errors can compound during closed-loop execution and disrupt a trained imitation learning policy. We study whether the original expert demonstrations can improve robustness through test-time hidden-state steering without additional demonstrations or policy retraining. Our analysis of a trained policy finds that expert hidden-state variation is concentrated in relatively few dimensions, while the directions capturing this variation differ across regions. Motivated by this geometry, we propose Diffeomorphic Hidden-State Steering (DHS), which learns an invertible nonlinear coordinate map from expert hidden states obtained by replaying the demonstrations through the frozen policy. DHS attenuates coordinates with low expert variance more strongly than those with high variance, then maps the modified coordinates back to hidden-state space. This allows steering directions to vary with the current hidden state while retaining every coordinate. Across four simulated manipulation tasks, DHS improves mean success under observation noise over the unmodified Transformer policies. Controlled comparisons yield higher mean success with learned nonlinear coordinates than with linear coordinates under both hard projection and full-rank rescaling. Combining DHS with action smoothing improves mean success rate under observation noise in certain tasks compared with action smoothing alone.