Research map / Vision-language and multimodal / Robot action and navigation
TwinVLA: Data-Efficient Bimanual Manipulation with Twin Single-Arm Vision-Language-Action Models
Hokyun Im, Euijin Jeong, Andrey Kolobov, Jianlong Fu, Youngwoon Lee
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 joint attention between the twin VLMs (ablation) | TwinVLA · real-world and Tabletop-Sim Easy | success rate: -27.0% real, -4.0% sim when removed | Improved |
| add attention re-weighting (ablation) | TwinVLA · real-world and Tabletop-Sim Easy | success rate: -1.1% sim, -4.0% real when removed; initial loss +40% | Improved |
| add MoE integration for shared inputs (ablation) | TwinVLA · real-world and Tabletop-Sim Easy | success rate: -1.1% sim, -5.0% real when removed; VRAM +21% | Improved |
| add single-arm OXE pretraining (ablation) | TwinVLA · real-world and Tabletop-Sim Easy | success rate: -4.6% sim, -46.0% real when from scratch | Improved |
| twin structure instead of monolithic RDT-1B | TwinVLA · real world, simulation, language following (avg) | success rate: +26.0% real, +5.0% sim, +21.8% language | Improved |
| use TwinVLA instead of RDT-1B | RoboTwin 2.0 Hard · RoboTwin 2.0 Hard | success rate: -3.72% vs RDT-1B | Hurt |
| use TwinVLA instead of RDT-1B | RoboTwin 2.0 Easy · RoboTwin 2.0 Easy | success rate: +7.48% | Improved |
| use TwinVLA instead of RDT-1B | Tabletop-Sim Hard · Tabletop-Sim Hard | success rate: +3.3% | Improved |
| TwinVLA vs pi0 under distractors | Fold towel (real) · Fold towel with distractors | success rate (%): 25.0 vs pi0 60.0 | Hurt |
| pretrain SingleVLA on OXE before LIBERO fine-tuning (ablation) | SingleVLA · LIBERO avg | success rate (%): 74.0 -> 86.0 | Improved |
Closest papers by meaning
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- ForceVLA: Enhancing VLA Models with a Force-aware MoE for Contact-rich Manipulation (NeurIPS 2025 · similarity 0.90)
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.89)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.89)
- BridgeVLA: Input-Output Alignment for Efficient 3D Manipulation Learning with Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation (ICLR 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.
