VFlash, 시각 계획으로 로봇 제어 속도 25배 향상
VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
- 매체
- Hugging Face
- 게시
- 중요도
- 4/5
- 화제 지수
- 22
한국어 요약
로봇과 자율주행 등 물리 AI 작업을 위해 시각 계획(점, 경계 상자, 궤적)을 활용하는 VFlash 기법을 제안했다. Flash Primitive Decoding을 통해 시각 원시 요소의 좌표를 한 번에 예측해 3D 점 동작 예측에서 약 25배 빠른 생성을 달성했다. 로봇 행동 모델이 VFlash 계획으로 조정될 때 MolmoSpaces 작업 묶음에서 평균 70.2% 성공률을 얻었고, 기존 방식은 62.6%였다.
같은 소식, 다른 매체 · 2곳
원문 발췌 · English
Models for physical AI tasks, such as robotics and autonomous driving, need not reason solely in language. They can use visual plans--points, bounding boxes, and trajectories--to represent spatial goals and anticipate motion. However, autoregressive vision-language models generate these plans one coordinate token at a time, making richer predictions increasingly costly. We introduce VFlash, a method for fast, accurate visual planning through Flash Primitive Decoding. VFlash exploits the known structure of visual primitives to predict their coordinates jointly, using a lightweight drafter that generates an entire primitive in one forward pass. The layout adapts to the primitive's type, dimensionality, and length, supporting compact boxes and long, variable-length trajectories. Crucially, spatial accuracy need not require reproducing an autoregressive model's exact token sequence. VFlash therefore trains on annotated primitives and directly accepts structurally valid predictions without target-token agreement. Across the evaluated 2D and 3D robot trajectory tasks, VFlash consistently reduces prediction errors relative to autoregressive decoding. Preliminary single-rollout measurements indicate approximately 25x faster generation for 3D point motion forecasting. A world action model conditioned on VFlash plans achieves 70.2% average success across four MolmoSpaces task suites, compared with 62.6% for the baseline.