RoboBench Benchmark for MLLM Robotics Evaluation
English multimodal test suite for checking how well vision-language models reason in robotics scenarios.
- Robotics
- Visual question answering
- English
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
English multimodal test suite for checking how well vision-language models reason in robotics scenarios.
Large-scale set of 1 to 10 million examples alternating images, text, and actions for training embodied AI.
Official set of tasks from AtCoder contests for evaluating AI systems in algorithmic programming with scoring.
A multimodal corpus of trajectories and action reasoning for training robotic manipulator control models.
Panda robot manipulation episodes: 1,693 episodes, 273k frames, and 40 tasks for training robotics policies.
8.43 TB of synchronized sensor data for autonomous driving and robotics.
A large dataset of human action videos and images for motion modeling and robot pre-training.
A large Physical AI dataset with 3D room scans and egocentric recordings in a unified metric system.
A 700+ million token corpus for training compact models in step-by-step task analysis and web search.
Over half a million vector graphic comparisons of 75 models across text prompts with 3.4 million human votes.
223 simulation-ready household, kitchen, and workshop objects in OpenUSD format with physics and semantics.
CAD assets and scenes with configured physics for home robot simulation.
1,000 interior 3D scenes in USDZ format for simulation and robotics in NVIDIA Isaac Sim.
High-precision collision meshes for 1,000 indoor scenes for physically accurate navigation and interaction.
A large-scale human motion dataset extracted from web video with detailed descriptions for robotics and animation generation.
13k images from OpenAI 4o with over 200k real human ratings for training preference models.
160 successful Gemini agent trajectories on OSWorld tasks with steps and 1.0 scores.
FPV drone simulator gameplay recordings with timestamped controller states.
A dataset of human pairwise evaluations (2,161 rows) for banana images from 11 generative models.