Research map / Reinforcement learning / Visual and representation RL
Sample from What You See: Visuomotor Policy Learning via Diffusion Bridge with Observation-Embedded Stochastic Differential Equation
Zhaoyang Liu, Mokai Pan, Zhongyi Wang, Kaizhen Zhu, Haotao Lu, Haipeng Zhang, Jingya Wang, Ye Shi
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 diffusion bridge head with standard diffusion (ablation) | BridgePolicy · Adroit Door | success rate: 0.68 -> 0.84 | Improved |
| replace regression head with diffusion bridge (ablation) | BridgePolicy · Adroit Door | success rate: 0.00 -> 0.84 | Improved |
| use diffusion bridge instead of diffusion (ablation) | BridgePolicy · Adroit Pen | success rate: 0.43 -> 0.58 | Improved |
| use cross-attention instead of concatenation for fusion (ablation) | BridgePolicy · Adroit Door | success rate: 0.59 -> 0.665 | Improved |
| use cross-attention instead of concatenation for fusion (ablation) | BridgePolicy · MetaWorld Handle-Pull | success rate: 0.55 -> 0.63 | Improved |
| add contrastive alignment loss (alpha 0.0 to 1.0) (ablation) | BridgePolicy · DexArt Laptop | success rate: 0.85 -> 0.91 | Improved |
| add alignment loss alpha 0.5 (ablation) | BridgePolicy · MetaWorld Coffee-Push | success rate: 0.91 -> 0.99 | Improved |
| raise alignment weight alpha from 1.0 to 5.0 (ablation) | BridgePolicy · MetaWorld Pick-Place-Wall | success rate: 0.96 -> 0.83 | Hurt |
| sample FlowPolicy with 10 steps instead of 1 (ablation) | FlowPolicy · 7 simulation tasks (avg) | success rate: 0.64 -> 0.59 | Hurt |
| L1 objective instead of L2 for diffusion bridge training (ablation) | BridgePolicy · Adroit (3 tasks avg) | success rate: 0.80 -> 0.81 | No clear effect |
| increase steady variance lambda from 30 to 50 (gamma 10^7) (ablation) | BridgePolicy · Adroit (3 tasks avg) | success rate: 0.77 -> 0.87 | Improved |
| increase penalty gamma from 10^5 to 10^7 (lambda 50) (ablation) | BridgePolicy · Adroit (3 tasks avg) | success rate: 0.83 -> 0.87 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FlowPolicy — FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025) · better on 7 of 8 shared results
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
- Prior Does Matter: Visual Navigation via Denoising Diffusion Bridge Models (CVPR 2025 · similarity 0.91)
- One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation (ICML 2025 · similarity 0.90)
- Two-Steps Diffusion Policy for Robotic Manipulation via Genetic Denoising (NeurIPS 2025 · similarity 0.90)
- Latent Diffusion Planning for Imitation Learning (ICML 2025 · similarity 0.90)
- Prediction with Action: Visual Policy Learning via Joint Denoising Process (NeurIPS 2024 · similarity 0.89)
- FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency (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.
