BEACON: Cross-Domain Co-Training of Generative Robot Policies via Best-Effort Adaptation

2026-05-08

AI Summary

This work addresses the challenge of effectively leveraging abundant source-domain data to train generalizable robotic policies when only a limited number of target-domain demonstrations are available. The authors propose a discrepancy-aware importance reweighting co-training method that jointly learns a diffusion-based visuomotor policy and sample-level weights for source-domain data. This approach implicitly aligns cross-domain features without requiring explicit alignment, while enabling balanced utilization of multiple source domains and scalable adaptation to high-dimensional sequential policies. Experimental results demonstrate that the proposed method significantly outperforms baselines—including those using only target data, fixed-ratio source–target mixing, or explicit feature alignment—across simulation-to-simulation, simulation-to-real, and multi-source manipulation tasks, achieving markedly improved policy robustness and data efficiency.

Abstract

We introduce BEACON--Best-Effort Adaptation for Cross-Domain Co-Training--a theory-driven framework for training generative robot policies with abundant source demonstrations and limited target demonstrations. BEACON casts cross-domain co-training as a discrepancy-aware importance-reweighting problem, jointly learning a diffusion-based visuomotor policy and per-sample source weights that minimize an objective informed by target-domain generalization guarantees. To make best-effort adaptation practical for high-dimensional sequence policies, we develop scalable instance-level discrepancy estimators, stochastic alternating updates for policy and weights, and a multi-source extension that balances heterogeneous source domains. Across sim-to-sim, sim-to-real, and multi-source manipulation settings, BEACON improves robustness and data efficiency over target-only, fixed-ratio co-training, and feature-alignment baselines. Importantly, even without an explicit alignment objective, BEACON achieves feature alignment as an implicit result of discrepancy-aware cross-domain co-training.

Problem

• cross-domain adaptation
• robot policy learning
• generative policies
• domain discrepancy
• data efficiency

Innovation

• cross-domain co-training
• discrepancy-aware reweighting
• diffusion-based policy
• instance-level discrepancy estimation
• implicit feature alignment