Skip to content
Robotics

trossen-bin-picking-act-cup-real

robot control policy for cup picking

2 min readupdated

A compact 50M-parameter model from the Kite team built on the ACT architecture (Action Chunking with Transformers, an algorithm that predicts sequences of robot actions at once) for the bimanual WidowX AI robotic station. It takes feeds from four cameras alongside the current positions of 14 joints, outputting target coordinates for all 14 joints 100 steps ahead. Trained on real-world robot motion recordings from Trossen Robotics' open bin_picking dataset, this policy outperforms counterparts trained solely in simulation. It is built for robotics developers and researchers working with the LeRobot framework.

Key features

  • Multi-sensor input: simultaneously processes streams from four cameras and the current positions of the station's 14 actuators.
  • Action chunking: generates a motion trajectory 100 steps ahead across 14 joints in a single forward pass.
  • Trained on real-world data: weights were trained on 18 real recordings of cup picking (7,318 frames) rather than synthetic data.
  • Task-specific design: takes no text prompts and is strictly tuned to grasp a red cup and place it into a clear bin.

What you can do

  • Run in simulation: test bimanual WidowX AI control under varying lighting conditions and with distractor objects on the table.
  • Benchmark training approaches: use the checkpoint as a baseline when comparing policies trained on real trajectories versus simulated trossen-bin-picking-sim datasets.
  • Fine-tune for custom tasks: use these weights as a starting point in LeRobot 0.6 and newer for object manipulation experiments.

Example

from lerobot.policies.act.modeling_act import ACTPolicy

policy = ACTPolicy.from_pretrained("kiteml/trossen-bin-picking-act-cup-real")
Read moreDetails
Simulation results: 33 successful picks out of 80 (10/20 under normal lighting, 8/20 in dim light, 8/20 in warm light, and 7/20 with extra objects on the table), beating the best simulation-trained model (26/80).
Training: 30,000 steps with a batch size of 8 on a single A100 GPU took roughly 1 hour and 50 minutes.
Limitations: evaluated only in simulation (not yet tested on physical hardware) and only succeeds when grasping with the left arm.
License: Apache-2.0.
Discuss Telegram 0 20

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.

    # kiteml/trossen-bin-picking-act-cup-real
    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 kiteml/trossen-bin-picking-act-cup-real --local-dir ~/models/trossen-bin-picking-act-cup-real
    # 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 trossen-bin-picking-act-cup-real (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

kiteml/trossen-bin-picking-act-cup-real
Author
kiteml
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

200

Ask AI about this model

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

Comments