FOCA: Future-Oriented Conditioning for Data-Efficient Vision-Language-Action Adaptation

2026-06-18

AI Summary

Existing vision-language-action (VLA) models exhibit limited generalization in few-shot imitation learning. This work proposes a future-conditioned framework that enables long-horizon reasoning without pixel-level reconstruction by explicitly predicting task-relevant future interaction embeddings and implicitly aligning goal observations in a latent space. The approach introduces, for the first time, a future-oriented mechanism into few-shot VLA adaptation, allowing joint training using only action-free videos and interpretable as learning a value-like future-conditioned representation. Experiments demonstrate that the method achieves a 95.7% success rate on the LIBERO benchmark with merely 20 demonstrations, yields 7–12% absolute improvements on RoboCasa, and attains up to a 26% absolute gain in real-world robotic tasks, substantially advancing the state of the art in few-shot VLA adaptation.

Abstract

Vision-Language-Action (VLA) models enable general-purpose robotic control via large-scale multimodal pretraining, yet their effectiveness under few-shot imitation learning remains limited. We conduct a systematic stress test of state-of-the-art VLA models and show that performance degrades sharply as demonstrations are reduced, revealing a key weakness of existing adaptation strategies. To address this, we introduce FOCA, a future-oriented conditioning framework for data-efficient VLA adaptation. FOCA combines explicit prediction of task-grounded future interaction embeddings with implicit alignment to future goal observations, enabling long-horizon reasoning in latent space without pixel-level prediction. This formulation naturally supports action-free co-training with synthetic videos from video world models and can be interpreted as learning a future-conditioned value-like representation. Extensive experiments demonstrate FOCA achieves 95.7% success with 20 demonstrations on LIBERO, improves 7-12% on RoboCasa, and delivers up to 26% absolute gains on real robots, establishing a new state of the art in few-shot VLA adaptation.

Problem

• Vision-Language-Action
• few-shot imitation learning
• data-efficient adaptation
• robotic control
• multimodal pretraining

Innovation

• future-oriented conditioning
• vision-language-action adaptation
• data-efficient imitation learning
• latent-space reasoning
• action-free co-training