Instructions to use simaai/Pi0.5-SO101-PickPlace with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use simaai/Pi0.5-SO101-PickPlace with LeRobot:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Pi0.5 SO-101 Pick-and-Place
A Pi0.5 policy fine-tuned to place a grey eraser in a white basket and compiled for the SiMa.ai Modalix DevKit.
| Property | Value |
|---|---|
| Base model | LeRobot Pi0.5 |
| Training checkpoint | Step 20,000 |
| Output | 50 actions x 6 joints |
| Denoising steps | 10 |
| Precision | W8A16 with BF16 attention |
| Compiled graphs | 64 |
| Toolchain | Model Compiler 3.0.0, Palette SDK 3.0.0, LLiMa 0.6.0 |
| Bundle ID | pi05-so101-grey-eraser-pickplace-20k-w8a16-fb509ec9 |
This repository contains the deployable compiled bundle, reference fixtures, tokenizer, embedding table, and SO-101 runtime assets. It does not contain the training checkpoint or dataset.
Requirements
- Modalix DevKit running Palette SDK 3.0.0 and LLiMa 0.6.0
- PoLiMa with SDK 3.0 and Pi0.5 support
- SO-101 follower with two USB cameras matching the training views
See the PoLiMa installation guide and SO-101 setup guide.
Download and deploy
hf download simaai/Pi0.5-SO101-PickPlace --local-dir Pi0.5-SO101-PickPlace
cd Pi0.5-SO101-PickPlace
sha256sum --check SHA256SUMS
read -r -p "DevKit login (username@address): " devkit_login
polima deploy --bundle ./bundle --devkit "$devkit_login" --start
Deployment starts the inference server but does not start robot motion.
Validate inference
polima run --bundle ./bundle --devkit "$devkit_login" --fixture --repeat 20
The packaged eight-case fixture passed compiled-device fidelity validation. On the SDK 3.0 reference DevKit, 20 measured plans after three warm-ups had a median model latency of 334 ms and cosine similarity 0.999920 against the reference output. This fixture test does not move the robot or measure physical task success.
Training data
The policy used 769 demonstrations from SO-101 Grey Eraser Pick-and-Place. The complete dataset contains 961 episodes and 272,853 frames.
Limitations
This policy is specific to the reference robot, object, basket, and camera views. Changes to geometry, framing, lighting, or calibration can affect behavior. A new instruction can select behavior learned during fine-tuning; it does not teach a new task at runtime.
License
This model is derived from lerobot/pi05_base, whose model card specifies the
Gemma license. Review those terms
before use or redistribution. PoLiMa source code has separate Apache-2.0 terms.