VLA Robotics Benchmark Index
Primary-source index of vision-language-action models, manipulation benchmarks, and robot datasets
A vision-language-action model maps camera images and a language instruction to robot motor commands. This index records public VLA models, manipulation benchmarks, and training datasets, copying figures only from author papers and project pages retrieved on 19 October 2026. Records include OpenVLA, RT-2, π0, Octo, LIBERO, and Open X-Embodiment.
60-Second Technical Overview & Benchmark Walkthrough
Verified VLA Robotics Benchmark Index Matrix (14 of 14 Tracked)
Backed by SQLite warehouse (warehouse/warehouse.db) · Last 30 days sensor verified
| Entity / Implementation | Organization | Status | Key Technical Metric | Governance / License | Verification |
|---|---|---|---|---|---|
| OpenVLA Model 7B open VLA trained on 970k robot demonstrations, using Llama 2 with fused DINOv2 and SigLIP features. | Stanford / UC Berkeley | arXiv v3 | 7B; 970k demos | Open code and checkpoints | verified |
| OpenVLA-OFT Model Stanford fine-tune of OpenVLA with 97.1% average success on four LIBERO suites and 26x faster actions. | Stanford University | Project release | 97.1% LIBERO avg | Stanford project release | verified |
| RT-2 Model Google DeepMind VLA that writes robot actions as text tokens, at up to 55B parameters over 6k trials. | Google DeepMind | arXiv 2023 | Up to 55B; 6k trials | Google DeepMind research | verified |
| RT-1 Model Robotics at Google transformer reporting 97% success on 700 instructions from 130k episodes on 13 robots. | Robotics at Google | RSS 2023 | 97% on 700 instructions | Robotics at Google | verified |
| Octo Model Open transformer policy trained on 800k Open X-Embodiment trajectories and tested across 9 robot platforms. | UC Berkeley | arXiv 2024 | 800k trajectories | Open pipeline, weights, and data | verified |
| π0 Model Flow-matching VLA from Physical Intelligence, pretrained on 7 robot configurations and 68 tasks at up to 50 Hz. | Physical Intelligence | arXiv preprint | 7 robots; 68 tasks | Physical Intelligence research | verified |
| SmolVLA Model Hugging Face VLA with 450M parameters and a flow-matching action expert, released with code, weights, and data. | Hugging Face | arXiv 2025 | 450M parameters | Code, weights, and data released | verified |
| GR00T N1 Model NVIDIA humanoid VLA. Public GR00T-N1-2B has 2.2B parameters, and the action module runs at 120 Hz. | NVIDIA | Open foundation model | 2.2B parameters | Publicly released model | verified |
| RDT-1B Model Tsinghua diffusion transformer with 1.2B parameters, pretrained on 46 datasets and more than 1M episodes. | Tsinghua University | Project release | 1.2B; 1M+ episodes | Tsinghua research release | verified |
| Open X-Embodiment Dataset Cross-lab robot dataset with 1M+ trajectories, 22 embodiments, and 60 source datasets stored as RLDS. | Open X-Embodiment Collaboration | Public dataset | 1M+ traj.; 22 robots | RLDS research dataset | verified |
| LIBERO Benchmark Lifelong manipulation benchmark with 130 language-conditioned tasks in Spatial, Object, Goal, and 100-task suites. | Bo Liu, Yuke Zhu, Peter Stone et al. | Public benchmark | 130 tasks; 4 suites | Open benchmark | verified |
| CALVIN Benchmark Freiburg long-horizon benchmark: 4 indoor environments, about 24 hours of play data, and 20K language directives. | University of Freiburg | Public benchmark | 4 envs; 20K directives | Open-source simulated benchmark | verified |
| SimplerEnv Benchmark Real-to-sim policy eval with over 1500 paired sim-and-real trials across two embodiments and eight task families. | Li, Hsu, Gu, Mees, Pertsch et al. | PMLR v270 | 1500+ paired evals | Open-sourced environments | verified |
| DROID Dataset In-the-wild Franka dataset: 76k trajectories, 350 hours, 564 scenes, and 86 tasks across 13 institutions. | DROID collaboration | Public dataset | 76k traj.; 350 hours | Open dataset and hardware guide | verified |
Featured Head-to-Head Comparisons
Every pair of the 14 records (91 total pairs) has a static comparison page. These 6 are the highest-intent matches.
Frequently Asked Questions
What is a vision-language-action model?
RT-2 uses the term for a vision-language model co-fine-tuned so robot actions are written as text tokens and trained with web vision-language tasks. At run time those tokens are turned back into motor commands.
How large is OpenVLA, and what was it trained on?
The OpenVLA paper describes a 7B-parameter model trained on 970k real-world robot demonstrations. It uses a Llama 2 backbone with visual features fused from DINOv2 and SigLIP.
What does the LIBERO benchmark contain?
LIBERO fixes 130 language-conditioned manipulation tasks in four suites: LIBERO-Spatial, LIBERO-Object, LIBERO-Goal, and LIBERO-100. The first three suites contain 10 tasks each, and LIBERO-100 is split into 90 short-horizon tasks and 10 long-horizon tasks.
How does π0 generate actions?
The π0 paper adds a flow-matching action expert on a pretrained vision-language model. The reported pretraining mix covers 7 robot configurations and 68 tasks, and the policy can run at up to 50 Hz.
What is in the Open X-Embodiment dataset?
The Open X-Embodiment paper reports 1M+ real-robot trajectories across 22 embodiments, pooled from 60 datasets and 34 research labs and stored in RLDS. The abstract also counts 527 skills.
How does vlabench.dev check a number?
Each row cites one author paper or project page retrieved on 2026-10-19. A figure is marked verified when that page states it. Model names, URLs, and scores are not filled in from memory.