DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation

Jusuk Lee, Seungjae Lee, Jonghun Shin, Hoseong Jung, Sungha Kim, Daesol Cho, H. Jin Kim, Jia-Bin Huang, Furong Huang

arXiv:2605.30350 · 2026-05-29 공개 · arXiv · PDF

vla simulation-real-world dynamics-aware image-language-3d hyperspherical-space contrastive-objective visual-backbones motion-understanding

Abstract

Robot manipulation critically depends on perception that preserves the action-relevant aspects of a scene. Yet most robot learning pipelines are built upon visual encoders pre-trained for static recognition or vision-language alignment, leaving motion understanding to downstream policies. We introduce DynaFLIP, a dynamics-aware multimodal pre-training framework that pushes motion understanding upstream into perception. We construct image-language-3D flow triplets from heterogeneous human and robot videos, and use these triplets as training-time supervision to shape an image-only encoder. Our key idea is to encourage the three modalities to span a small simplex volume in the shared hyperspherical space -- a smaller simplex volume indicating stronger alignment. To avoid the geometric ambiguity and trivial collapse of naive volume minimization, we combine simplex-volume minimization with a cosine regularizer and a contrastive objective. Our analyses show that DynaFLIP focuses on control-relevant regions critical for manipulation. The resulting dynamics-aware representations serve as reusable visual backbones and consistently outperform baselines across diverse downstream policies, including VLAs. We validate this across diverse simulation and real-world setups, with gains reaching +22.5% under out-of-distribution scenarios. Our results suggest that robot generalization improves when visual representations are trained to encode not just what is present, but how the world changes under action.

한국어 요약

📋 한 줄 요약

**[로봇 perception / Dynamics-Aware Pre-training]** DynaFLIP가 image-language-3D flow triplet을 simplex volume minimization·cosine regularizer·contrastive objective로 정렬, 동작 이해를 perception 단으로 push — 다양 downstream policy(VLA 포함)에서 OOD +22.5% 향상.

🎯 핵심 기여도

💡 핵심 아이디어

로봇 generalization은 visual 표현이 무엇이 있는지(what)뿐 아니라 행동에 따른 변화(how)도 인코딩하도록 학습될 때 개선된다 — image-language-3D flow를 작은 simplex volume에 span하도록 정렬하면 dynamics를 perception에 upstream으로 통합할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 로봇 perception에 dynamics 이해를 upstream으로 통합하는 새 패러다임, simplex volume 정렬이라는 일반 가능 정렬 원리, OOD 향상으로 generalization 검증. **한계**: 3D flow 데이터 확보 부담(label 비용), simplex volume의 hyperparameter 조정, 매우 dexterous·long-horizon task 일반화는 후속.

🚀 실용적 활용