Robotics · 2026-11-04

Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution

Zhanyi Sun, Shuran Song

arXiv:2508.05941PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

Latent Policy Barrier (LPB) improves visuomotor policy robustness under covariate shift by treating latent embeddings of expert demonstrations as an implicit barrier separating in-distribution from out-of-distribution states. It trains a base diffusion policy on expert data and an action-conditioned latent dynamics model on both expert and rollout data; at inference, it steers actions using latent-space optimization to stay near the expert latent manifold.

Why it matters

The authors address covariate shift in behavior cloning, where small deviations can compound into failures. LPB decouples precise imitation from OOD recovery into separate training modules and performs inference-time steering without additional human correction, rewards, or task rewards. The work claims improved robustness and sample efficiency in both simulation and real-world manipulation.

Method

  • Train a base diffusion policy on limited high-quality expert demonstrations, and train a visual latent dynamics model on both expert demos and policy rollout data.
  • Define an out-of-distribution latent score δ(z) by nearest-neighbor distance from the expert latent manifold.
  • At inference, if δ(z_t) exceeds a threshold τ, refine denoising using gradients from the latent dynamics model to minimize latent distance to nearest expert latent states.
Abstract (from arXiv)

Visuomotor policies trained via behavior cloning are vulnerable to covariate shift, where small deviations from expert trajectories can compound into failure. Common strategies to mitigate this issue involve expanding the training distribution through human-in-the-loop corrections or synthetic data augmentation. However, these approaches are often labor-intensive, rely on strong task assumptions, or compromise the quality of imitation. We introduce Latent Policy Barrier, a framework for robust visuomotor policy learning. Inspired by Control Barrier Functions, LPB treats the latent embeddings of expert demonstrations as an implicit barrier separating safe, in-distribution states from unsafe, out-of-distribution (OOD) ones. Our approach decouples the role of precise expert imitation and OOD recovery into two separate modules: a base diffusion policy solely on expert data, and a dynamics model trained on both expert and suboptimal policy rollout data. At inference time, the dynamics model predicts future latent states and optimizes them to stay within the expert distribution. Both simulated and real-world experiments show that LPB improves both policy robustness and data efficiency, enabling reliable manipulation from limited expert data and without additional human correction or annotation.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2025-08-08 · Indexed by Pub-AI: 2026-11-04 UTC

preprint

Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution

Zhanyi Sun, Shuran Song

Preprint posted: 2025-08-08 · Original record ↗

arxiv
2508.05941
Explicit source versions (1)
  • v2 · 2026-07-01
  1. Explicit source version detected · v2

    Source date: 2026-07-01 · First detected: 2026-11-05 UTC · Historical baseline

    Source evidence (1)
  2. Preprint available

    Source date: 2025-08-08 · First detected: 2026-11-05 UTC · Historical baseline

    Source evidence (1)
Abstract · arxiv

Visuomotor policies trained via behavior cloning are vulnerable to covariate shift, where small deviations from expert trajectories can compound into failure. Common strategies to mitigate this issue involve expanding the training distribution through human-in-the-loop corrections or synthetic data augmentation. However, these approaches are often labor-intensive, rely on strong task assumptions, or compromise the quality of imitation. We introduce Latent Policy Barrier, a framework for robust visuomotor policy learning. Inspired by Control Barrier Functions, LPB treats the latent embeddings of expert demonstrations as an implicit barrier separating safe, in-distribution states from unsafe, out-of-distribution (OOD) ones. Our approach decouples the role of precise expert imitation and OOD recovery into two separate modules: a base diffusion policy solely on expert data, and a dynamics model trained on both expert and suboptimal policy rollout data. At inference time, the dynamics model predicts future latent states and optimizes them to stay within the expert distribution. Both simulated and real-world experiments show that LPB improves both policy robustness and data efficiency, enabling reliable manipulation from limited expert data and without additional human correction or annotation.