This site is an independent community hub, not affiliated with Hugging Face. Official repo: https://github.com/huggingface/lerobot
How to Choose LeRobot Models and Datasets for Fast Reproduction?
Direct Answer: Filter models by hardware first, then choose the algorithm type (Diffusion/ACT/TD-MPC2) best suited for your task.
| Algorithm | Pros | Cons | Best For |
|---|---|---|---|
| Diffusion Policy | Smoother actions | Heavy inference | Fine grasping |
| ACT | Stable to train | Data sensitive | Dual-arm tasks |
| TD-MPC2 | Strong dynamic control | Complex tuning | Humanoids/Walking |
| Hardware | Recommended Use |
|---|---|
| SO-100 | Beginner grasp loop |
| Aloha | Fine dual-arm operation |
| Unitree | Dynamic tasks & mobility |
| Reproduction Method | Pros | Risks | Recommendation |
|---|---|---|---|
| Download & Inference | Fastest | Environment diff | Validate motion first |
| Fine-tune | Fast convergence | Transfer failure | Similar tasks |
| Train from scratch | Most controllable | High cost | Very different tasks |
lerobot/diffusion_pusht
Diffusion PolicyDiffusion Policy model trained in the PushT environment (gym-pusht) to push a T-shaped block to a target position. Official reference implementation of Diffusion Policy in LeRobot.
lerobot/act_aloha_sim_transfer_cube_human
ACTACT (Action Chunking with Transformers) policy trained in the Aloha simulation environment for cube transfer tasks, using human teleoperation demonstrations.
lerobot/act_aloha_sim_insertion_human
ACTACT policy trained in the Aloha simulation environment for fine-grained insertion tasks (AlohaInsertion), using human teleoperation demonstration data.
Want to Share Your Model?
If you have trained an excellent model or compiled a high-quality dataset, welcome to submit it to us. Authors featured on the curated list will receive a special lerobot.tech community badge and hardware discounts.
Submit Model Card