Skip to content
Robotics

act_so101_santa_ducky_v1_toponly

Robot control model for toy grasping task

2 min readupdated

A new control policy model created using the LeRobot library based on the ACT architecture (Action Chunking with Transformers — an imitation learning method predicting short action sequences instead of single steps). The model is trained for a so_follower type robot and takes the robot's state and an image from the top camera as input to predict control actions. It solves a specific practical task of grasping and moving a toy object onto a plate. The model will be useful for researchers and practitioners working with robotic arms and the LeRobot library.

Key features

  • Built on the ACT architecture for sequential robot action prediction
  • Uses data from the top camera and the manipulator's current state
  • Trained on a demonstration dataset of 58 episodes (15,558 frames at 30 FPS)
  • Trained using the AdamW optimizer for 40,000 steps

What you can do

  • Run the pre-trained model on a compatible so_follower robot to perform the object moving task
  • Use the demonstration dataset for fine-tuning your own manipulation models
  • Configure the launch settings via command line using LeRobot utilities

Example

Running the policy on your robot:

lerobot-rollout \
  --strategy.type=base \
  --robot.type=so_follower \
  --robot.port= \
  --robot.cameras="{ : {type: opencv, index_or_path: , width: 640, height: 480, fps: 30}}" \
  --policy.path=boinkie/act_so101_santa_ducky_v1_toponly \
  --task="Pick up the santa ducky and place it in the plate" \
  --duration=60
Read moreDetails
Task: grasp the duck santa and place it on a plate.
Robot type: so_follower with wrist and top cameras.
Training parameters: Batch size 8, Learning rate 1e-05, Seed 1000, LeRobot version 0.6.2.
Data volume: 58 episodes, 15,558 frames, 30 FPS frequency.
Discuss Telegram 0 1

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 0.2 GB · ready files0.7 GB
  • BF16weights 0.1 GB0.6 GB
  • FP8 / INT8recommendedweights 0.0 GB0.6 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F32 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # boinkie/act_so101_santa_ducky_v1_toponly
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download boinkie/act_so101_santa_ducky_v1_toponly --local-dir ~/models/act_so101_santa_ducky_v1_toponly
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of act_so101_santa_ducky_v1_toponly (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

boinkie/act_so101_santa_ducky_v1_toponly
Author
boinkie
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

10

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments