Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Intention Distillation (INDI), a novel approach that enhances Vision-Language-Action (VLA) models by explicitly distilling behavior-level intent into action decoders. By leveraging a frozen teacher VLM to interpret multimodal inputs, INDI significantly improves action prediction performance across various benchmarks, achieving notable gains in both simulated and real-world tasks. The results demonstrate that incorporating a semantic understanding of behavior objectives leads to more effective action generation, with improvements in success rates and predictive organization.
Explicitly modeling the semantic objectives of robot actions can boost performance by over 20% in multimodal environments.
Vision-Language-Action (VLA) models can turn multimodal context into robot actions, but their action decoders are still trained largely by behavior cloning. This supervises which motor command was demonstrated while leaving implicit the local objective served by the behavior under the instruction. Future-based supervision enriches action learning with frames, latent observations, trajectories, or motion representations, but these signals capture particular realizations of what may happen rather than the shared semantic objective of the forthcoming behavior. We propose Intention Distillation (INDI), which distills behavior-level intent into the action decoder. During training, a frozen teacher VLM interprets a demonstrated segment from the current observation, instruction, coarse action summary, and corresponding execution video. From its standard inputs, the deployed VLA recovers the resulting multimodal intent representation at an intermediate decoder layer and uses it to organize action prediction together with representations of how the behavior unfolds and what it achieves. On SimplerEnv-Bridge, INDI improves GR00T-N1.7 from 64.3% to 84.7%, and on RoboCasa Kitchen it improves the controlled GR00T-N1.7 baseline from 64.1% to 70.3%, with consistent gains on π_{0.5} across both benchmarks. In real-world tasks, INDI improves average success from 62.0% to 68.7%, with gains of up to 12.0 pp on longer-horizon tasks. Further analyses show that the recovered latent is used by the decoder, captures behavior objective and execution progress, and organizes downstream predictions in an objective-dependent manner. These results show that action decoders benefit from explicitly modeling the semantic objective of the behavior they generate.