Research map / Vision-language and multimodal / Robot action and navigation
ForeAct: Steering Your VLA with Efficient Visual Foresight Planning
Zhuoyang Zhang, Shang Yang, Qinghao Hu, Luke J. Huang, James Hou, Yufei Sun, Yao Lu, Song Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add visual foresight planning (imagined future observation + subtask text) | pi0 + VLM · Real-world 11 tasks | average success rate (%): 57.1 -> 87.4 | Improved |
| add VLM subtask planning to pi0 | pi0 · Real-world 11 tasks | average success rate (%): 46.5 -> 57.1 | Improved |
| add foresight planning to pi0.5 | pi0.5 · Real-world tasks | average success rate (%): 70.3 -> 88.2 | Improved |
| add foresight planning to pi0.5 | pi0.5 · LIBERO average | success rate (%): 96.8 -> 97.5 | No clear effect |
| pretrain foresight generator on cross-embodiment data (ablation) | foresight image generator · Out-of-distribution tasks | fidelity: 0.00 -> 0.88 | Improved |
| pretrain foresight generator on cross-embodiment data (ablation) | foresight image generator · In-distribution tasks | fidelity: 0.18 -> 1.00 | Improved |
| use spatial text instruction instead of semantic text for pi0 (ablation) | pi0 · Pick_Tool | success rate (%): 20.0 -> 46.8 | Improved |
| replace spatial text with foresight goal image (ours) (ablation) | pi0 · Pick_Tool | success rate (%): 46.8 -> 93.4 | Improved |
| increase denoising steps for foresight generation beyond 8 (ablation) | foresight generator · image quality / latency | visual quality: 0.33 s at 8 steps | No clear effect |
| reduce denoising steps to 1-2 (ablation) | foresight generator · image quality | artifacts: 0.13s at 1 step | Hurt |
| use ForeAct with less training data (20% of Clean_Rubb data) | pi0 + ForeAct · Clean_Rubb | success rate (%): 79% at 20% data | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CoT-VLA — CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025) · better on 1 of 1 shared result
- CogVLA — CogVLA: Cognition-Aligned Vision-Language-Action Models via Instruction-Driven Routing & Sparsification (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CoFFT: Chain of Foresight-Focus Thought for Visual Language Models (NeurIPS 2025 · similarity 0.90)
- World–Value–Action Model: Implicit Planning for Vision–Language–Action Systems (NeurIPS 2026 · similarity 0.90)
- See Further, Think Deeper: Advancing VLM's Reasoning Ability with Low-level Visual Cues and Reflection (CVPR 2026 · similarity 0.89)
- FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving (NeurIPS 2025 · similarity 0.89)
- 3D-VLA: A 3D Vision-Language-Action Generative World Model (ICML 2024 · similarity 0.89)
- VLA-OS: Structuring and Dissecting Planning Representations and Paradigms in Vision-Language-Action Models (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
