Research map / Reinforcement learning / Visual and representation RL
CUBic: Coordinated Unified Bimanual Perception and Control Framework
Xingyu Wang, Pengxiang Ding, Jingkai Xu, Donglin Wang, Zhaoxin Fan
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 shared mapping for dual codebooks (ablation) | CUBic · RoboTwin | mean success rate (%): 40.1 -> 51.8 | Improved |
| add two-stage training (single-arm pre-training then dual-arm post-training) (ablation) | CUBic · RoboTwin | mean success rate (%): 42.1 -> 51.8 | Improved |
| add both shared mapping and two-stage training (ablation) | CUBic · RoboTwin | mean success rate (%): 32.6 -> 51.8 | Improved |
| add shared mapping without two-stage training (ablation) | CUBic · RoboTwin | mean success rate (%): 32.6 -> 42.1 | Improved |
| add two-stage training without shared mapping (ablation) | CUBic · RoboTwin | mean success rate (%): 32.6 -> 40.1 | Improved |
| add latent tokens (N=4) before quantization (ablation) | CUBic · RoboTwin | mean success rate (%): 0.0 -> 51.8 | Improved |
| use more latent tokens and larger codebook (N=8, K=512 vs N=4, K=256) (ablation) | CUBic · RoboTwin | mean success rate (%): 51.8 -> 40.2 | Hurt |
| replace DP with CUBic | Diffusion Policy · RoboTwin (7 tasks average) | success rate (%): 38.5 -> 51.8 | Improved |
| replace DP3 (with point clouds) with RGB-only CUBic | DP3 · RoboTwin (7 tasks average) | success rate (%): 39.8 -> 51.8 | Improved |
| replace DP with CUBic | Diffusion Policy · Dual Bottles Pick (Easy) | success rate (%): 85.7 -> 84.3 | No clear effect |
| replace DP3 with CUBic | DP3 · Dual Shoes Place | success rate (%): 12.0 -> 10.0 | No clear effect |
| replace DP with CUBic on real robot in-domain | Diffusion Policy · Real-world in-domain | step-wise score: 19.5 -> 43.1 | Improved |
Closest papers by meaning
- Rethinking Bimanual Robotic Manipulation: Learning with Decoupled Interaction Framework (ICCV 2025 · similarity 0.89)
- Unifying Perception and Action: A Hybrid-Modality Pipeline with Implicit Visual Chain-of-Thought for Robotic Action Generation (CVPR 2026 · similarity 0.88)
- Hierarchical Visuomotor Control of Humanoids (ICLR 2019 · similarity 0.87)
- SPIN: Simultaneous Perception Interaction and Navigation (CVPR 2024 · similarity 0.87)
- CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation (ICML 2026 · similarity 0.86)
- Closed-Loop Visuomotor Control with Generative Expectation for Robotic Manipulation (NeurIPS 2024 · similarity 0.86)
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.
