Reflective VLA: In-Context Action Consequences Make VLAs Generalize

2026-06-23

AI Summary

Existing vision-language-action (VLA) models are predominantly reactive and struggle to infer robot embodiment characteristics—such as camera geometry or actuation biases—from a single observation, limiting their generalization across environments. This work proposes Reflective VLA, the first approach to explicitly model the environmental impact of actions by treating action consequences as critical contextual information within observation–action–consequence triplets. The method employs a shared-attention architecture grounded in vision-language models, combined with blockwise causal masking to enable efficient parallel training while supporting real-time inference via KV caching. Experiments on LIBERO-Plus and its Hard variant demonstrate consistent improvements under distribution shifts, with average success rates increasing by 5.4 and 4.2 percentage points, respectively, without compromising performance on the original data distribution.

Abstract

Most vision-language-action (VLA) models are reactive: they predict the next action from the current instruction and observation, implicitly assuming that the current observation fully specifies the action-relevant state. In embodied control, however, embodiment-specific factors such as camera-to-robot geometry, robot calibration, or systematic actuation bias are often hard to identify from a single observation. As a result, reactive policies cannot reliably disambiguate these factors in general, overfitting to training environments and generalizing poorly at deployment. We propose Reflective VLA, which conditions each decision on a context of observation-action-consequence triplets. Each triplet records not only what the robot observed and executed, but also how the scene changed afterward, exposing the deployment-specific mapping from actions to observed effects. Architecturally, Reflective VLA routes all observation modalities through the VLM under shared attention, so the action expert reasons directly over past triplets and the current observation. A block-causal mask enables parallel multi-frame training without leakage and supports KV-cached real-time inference. On standard LIBERO and SimplerEnv-Bridge, Reflective VLA preserves strong in-distribution performance. Under distribution shift on LIBERO-Plus and the harder LIBERO-Plus-Hard, it improves average success rate by 5.4 and 4.2 percentage points over a matched reactive baseline. Ablations with a matched history-only baseline further show that action consequences -- rather than additional context length alone -- are the key to cross-environment generalization. Project page: https://lianqing11.github.io/reflective-vla-page/

Problem

• vision-language-action
• embodied control
• distribution shift
• generalization
• action consequences

Innovation

• Reflective VLA
• action consequences
• in-context learning
• embodied generalization
• vision-language-action