Research map / Vision-language and multimodal / Robot action and navigation
LatentVLA: Taming Latent Space for Generalizable and Long-Horizon Bimanual Manipulation
Junming Wang
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 |
|---|---|---|---|
| use continuous diffusion planner instead of VQ-VAE + autoregressive GPT (ablation) | LatentVLA · towel folding | success rate: 25.0 -> 58.0 | Improved |
| add absolute temporal encoding (ablation) | LatentVLA · towel folding | success rate: 31.0 -> 58.0 | Improved |
| add joint training with labeled robot data (ablation) | LatentVLA · towel folding | success rate: 35.0 -> 58.0 | Improved |
| add language-guided attention (ablation) | LatentVLA · towel folding | success rate: 39.0 -> 58.0 | Improved |
| add latent history conditioning (ablation) | LatentVLA · towel folding | success rate: 45.0 -> 58.0 | Improved |
| LatentVLA vs re-trained RDT-1B | RDT-1B · towel folding | success rate: 49.0 -> 58.0 | Improved |
| LatentVLA vs re-trained LAPA | LAPA · towel folding | success rate: 10.0 -> 58.0 | Improved |
| train LatentVLA from scratch on original 1.3M RDT-1B dataset | RDT-1B · real-world tasks | success rate: +5.6% | Improved |
| LatentVLA trained on all 2.5M trajectories vs RDT-1B fine-tuned on 1600 h | RDT-1B · real-world tasks | success rate: +7.2% | Improved |
| LatentVLA vs Moto on CALVIN | Moto · CALVIN (ABC->D) | average length: 3.10 -> 3.52 | Improved |
| LatentVLA vs LAPA on SIMPLER | LAPA · SIMPLER | success rate: 52.1 -> 65.7 | Improved |
| LatentVLA vs DP3 on RoboTwin pick apple messy | DP3 · RoboTwin 1.0 | success rate: 10.7 -> 33.7 | Improved |
Later papers that compared with this method
1 later accepted paper reports LatentVLA as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- ChainFlow-VLA: Causal Flow Planning with Vision-Language Models (NeurIPS 2026 · better on 1 of 1 shared result)
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.
- Moto — Moto: Latent Motion Token as the Bridging Language for Learning Robot Manipulation from Videos (ICCV 2025) · better on 2 of 2 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
- LaST$_{0}$: Latent Spatio-Temporal Chain-of-Thought for Robotic Vision-Language-Action Model (ICML 2026 · similarity 0.92)
- RotVLA: Rotational Latent Action for Vision-Language-Action Model (NeurIPS 2026 · similarity 0.91)
- Latent Diffusion Planning for Imitation Learning (ICML 2025 · similarity 0.90)
- LaST-R1: Reinforcing Robotic Manipulation via Adaptive Physical Latent Reasoning (NeurIPS 2026 · similarity 0.90)
- RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation (ICLR 2025 · similarity 0.90)
- Latent Action Pretraining from Videos (ICLR 2025 · similarity 0.90)
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.
