Driving · 2026-10-12
Evo-World: A Lightweight Vision-Language-Action Model with Future Motion Prediction via Optical-Flow Supervision
Zile Zhong, Yuxin Du, Tao Lin, Haoxiao Wang, Wei Dai, Binhua Zuo, Bo Zhao
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
Evo-World is a lightweight vision-language-action model that augments policy learning with future motion prediction via optical-flow supervision. It trains a motion representation module to predict task-relevant motion in latent space, using optical-flow renderings only during training, and preserves these predictive representations during policy learning via SIGReg and detached gradient paths. It reports strong simulation success and better real-world manipulation performance on an AgileX dual-arm robot while using 0.87B parameters.
Why it matters
Anticipating scene evolution can improve control, but action-only supervision may not enforce anticipation and future RGB prediction entangles motion with appearance. The authors propose optical-flow-supervised future motion prediction as a motion-focused predictive signal, and report simulation and real-world results showing improved success while keeping the model compact for deployment.
Method
- Train a motion representation module to predict instruction-conditioned latent motion using optical-flow renderings computed from demonstration videos; the target pathway is used only during training.
- Use an InternVL3-1B vision-language backbone and condition a flow-matching action expert with FiLM parameters derived from both current and predicted motion representations.
- Preserve predictive representations during policy learning with SIGReg (from LeJEPA) to discourage collapse and block action gradients into the motion representation module via stop-gradient.
Abstract (from arXiv)
Vision-language-action (VLA) models provide a unified framework for visual understanding, language grounding, and robot control, yet anticipating how the scene will evolve beyond the current observation can provide additional information for control. Action supervision alone does not explicitly encourage such anticipation, while future-image prediction entangles task-relevant motion with largely static appearance. To address these limitations, we introduce Evo-World, a lightweight VLA model that augments policy learning with future motion prediction through optical-flow supervision, thereby providing a task-relevant predictive signal focused on motion rather than appearance. We further design the training procedure to preserve these predictive representations during policy learning, regulating representation diversity while decoupling representation learning from action optimization. Across simulation benchmarks, Evo-World demonstrates strong performance, achieving a tier-averaged success rate of 91.1% on Meta-World MT50 , with a 13.0% relative improvement over the non-predictive baseline. On real-world AgileX robots, Evo-World outperforms the evaluated baselines across single-arm and bimanual manipulation tasks while using only 0.87B parameters and delivering lower GPU memory usage and higher inference efficiency. These results demonstrate that future motion prediction through optical-flow supervision provides an effective and lightweight approach to enhancing the predictive capabilities of VLA models.
Related papers
- WAVE-0: Bridging VLA Policies and WAMs with Lightweight Latent Future Prediction
- MotionWeave: Learning Motion-Centered Future Dynamics for Vision-Language-Action Policies
- Forward Dynamics Models
- TS-WAM: Temporal Latent Actions and World-Action Coupling for Vision-Language Robot Control
- PLaW-VLA: Predictive Latent World Modeling for Vision-Language-Action Policies
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-09 · Indexed by Pub-AI: 2026-10-12 UTC
Evo-World: A Lightweight Vision-Language-Action Model with Future Motion Prediction via Optical-Flow Supervision
Preprint posted: 2026-10-09 · Original record ↗
- arxiv
- 2610.12936
Explicit source versions (1)
- v1 · 2026-10-09
- Preprint available
Source date: 2026-10-09 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Vision-language-action (VLA) models provide a unified framework for visual understanding, language grounding, and robot control, yet anticipating how the scene will evolve beyond the current observation can provide additional information for control. Action supervision alone does not explicitly encourage such anticipation, while future-image prediction entangles task-relevant motion with largely static appearance. To address these limitations, we introduce Evo-World, a lightweight VLA model that augments policy learning with future motion prediction through optical-flow supervision, thereby providing a task-relevant predictive signal focused on motion rather than appearance. We further design the training procedure to preserve these predictive representations during policy learning, regulating representation diversity while decoupling representation learning from action optimization. Across simulation benchmarks, Evo-World demonstrates strong performance, achieving a tier-averaged success rate of 91.1% on Meta-World MT50 , with a 13.0% relative improvement over the non-predictive baseline. On real-world AgileX robots, Evo-World outperforms the evaluated baselines across single-arm and bimanual manipulation tasks while using only 0.87B parameters and delivering lower GPU memory usage and higher inference efficiency. These results demonstrate that future motion prediction through optical-flow supervision provides an effective and lightweight approach to enhancing the predictive capabilities of VLA models.