Research map / Vision-language and multimodal / Robot action and navigation
ContextFlow: In-Context Flow Matching for Robot Manipulation
Jian Ding, Xianjie DAI, Roei Herzig, Nussair Hroub, Jinjie Mai, Dengxin Dai, Bernard Ghanem, Mohamed Elhoseiny
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 |
|---|---|---|---|
| replace autoregressive action decoding with flow matching (both pi0-initialized) | ContextAR vs ContextFlow-Plain · LIBERO unseen configs | success rate (%): 53.5 -> 64.0 | Improved |
| add multimodal context compressors (ablation) | ContextFlow-Plain · LIBERO unseen configs | success rate (%): 64.0 -> 73.5 | Improved |
| add only image compressor (ablation) | ContextFlow-Plain · LIBERO unseen configs | success rate (%): 64.0 -> 68.0 (Spatial 55.0 -> 51.0, Object 73.0 -> 85.0) | Mixed |
| include proprio states and actions in context (ablation) | ContextFlow · LIBERO unseen configs | success rate (%): 29.0 -> 73.5 | Improved |
| include images in context (ablation) | ContextFlow · LIBERO unseen configs | success rate (%): 45.0 -> 73.5 | Improved |
| include text instruction in context (ablation) | ContextFlow · LIBERO unseen configs | success rate (%): 64.5 -> 73.5 | Improved |
| increase demo frames from 4 to 8 (ablation) | ContextFlow · LIBERO unseen configs | success rate (%): 70.5 -> 73.5 | Improved |
| increase states/actions from 128 to 256 (ablation) | ContextFlow · LIBERO unseen configs | success rate (%): 73.5 -> 66.5 | Hurt |
| use flow matching head instead of MLP L1 regression (ablation) | ContextFlow · LIBERO unseen configs | success rate (%): 58.0 -> 73.5 | Improved |
| use flow matching head instead of diffusion head (ablation) | ContextFlow · LIBERO unseen configs | success rate (%): 33.5 -> 73.5 | Improved |
| MLP L1 continuous parallel head vs autoregressive (ablation) | ContextAR · LIBERO unseen configs | success rate (%): 53.5 -> 58.0 | Improved |
| flow matching reduces cumulative error vs autoregression | ContextAR · Cumulative MSE on S2 task | MSE: 4.7 -> 3.0 | Improved |
Closest papers by meaning
- FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025 · similarity 0.89)
- Frequency‑Aware Flow Matching for Continuous and Consistent Robotic Action Generation (NeurIPS 2026 · similarity 0.88)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.88)
- Instant Policy: In-Context Imitation Learning via Graph Diffusion (ICLR 2025 · similarity 0.87)
- DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation (NeurIPS 2026 · similarity 0.87)
- LAFP: Preserving Latent Action Structure in Latent Policy Learning via Flow Matching (NeurIPS 2026 · similarity 0.87)
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.
