IntentVLA: Short-Horizon Intent Modeling for Aliased Robot Manipulation

2026-05-14

AI Summary

This work addresses the instability and action-chunk conflicts in vision-language-action (VLA) policies operating under partial observability, which arise from myopic intent discrepancies. To mitigate these issues, the authors propose a history-aware VLA framework that encodes recent visual observations into compact myopic intent representations to guide action-chunk generation, thereby enhancing policy consistency and stability. The study introduces myopic intent modeling as a novel mechanism to alleviate observation aliasing—a common challenge in partially observable environments—and presents AliasBench, a dedicated evaluation benchmark for this purpose. Extensive experiments across multiple simulated multi-task environments, including AliasBench, RoboTwin2, and SimplerEnv, demonstrate that the proposed method significantly outperforms strong existing VLA baselines, validating its effectiveness and robustness.

Abstract

Robot imitation data are often multimodal: similar visual-language observations may be followed by different action chunks because human demonstrators act with different short-horizon intents, task phases, or recent context. Existing frame-conditioned VLA policies infer each chunk from the current observation and instruction alone, so under partial observability they may resample different intents across adjacent replanning steps, leading to inter-chunk conflict and unstable execution. We introduce IntentVLA, a history-conditioned VLA framework that encodes recent visual observations into a compact short-horizon intent representation and uses it to condition chunk generation. We further introduce AliasBench, a 12-task ambiguity-aware benchmark on RoboTwin2 with matched training data and evaluation environments that isolate short-horizon observation aliasing. Across AliasBench, SimplerEnv, LIBERO, and RoboCasa, IntentVLA improves rollout stability and outperforms strong VLA baselines

Problem

• short-horizon intent
• observation aliasing
• robot imitation learning
• VLA policy
• multimodal demonstration

Innovation

• IntentVLA
• short-horizon intent
• visual-language-action policy
• observation aliasing
• history-conditioned modeling