Skip to content
Robotics

allros/act-franka-rk3588s

ACT checkpoint for robotics and deployment on RK3588S

2 min readupdated

This release includes a PyTorch checkpoint and a compiled NPU version for RK3588/RK3588S chips, built on the LeRobot ACT framework (Action Chunking with Transformers, an architecture for controlling robotic arms). Designed for imitation learning and MuJoCo simulations, the weights let you run inference on low-cost boards via built-in NPU optimization. The package contains both the original weights after 50,000 training steps and a ready-to-use RKNN file in FP16 format. It is useful for robotics developers, embedded systems engineers, and anyone testing robotic arm algorithms on Rockchip processors.

Key features

  • Two versions: original PyTorch weights (safetensors) and an optimized RKNN model (FP16, around 64.60 MiB) for RK3588 boards.
  • Hardware acceleration: native NPU support on RK3588S boards for HTTP service inference.
  • Ready-to-use preprocessors: input normalization configurations and parameters are already included.
  • Fast inference: average request processing time on the board is around 26–27 ms.

What you can do

  • Run a simulation: pair a PC running MuJoCo for rendering with an RK3588S board for NPU inference.
  • Integrate into robotics: use the RKNN model to control physical or simulated Franka robotic arms.
  • Run on PyTorch: load the weights via the lerobot library to experiment with the control policy.
  • Deploy an HTTP service: spin up an inference server on the board to send states and receive control actions remotely.

Example

from pathlib import Path
from huggingface_hub import snapshot_download
from lerobot.policies.act.configuration_act import ACTConfig
from lerobot.policies.act.modeling_act import ACTPolicy
from lerobot.policies.factory import make_pre_post_processors

path = Path(snapshot_download("allros/act-franka-rk3588s")) / "pytorch"
config = ACTConfig.from_pretrained(path)
config.device = "cpu"
config.pretrained_backbone_weights = None
policy = ACTPolicy.from_pretrained(path, config=config).cpu().eval()
Read moreDetails
• Architecture: LeRobot ACT (Action Chunking with Transformers).
• Performance: average call time on an RK3588S board is 26.06–27.55 ms.
• Formats: safetensors (PyTorch) and FP16 RKNN.
• Limitations: tested only at a fixed position in MuJoCo simulation; randomization and deployment on real hardware require additional steps.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.2 GB · ready files0.7 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.1 GB0.6 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 · BF16 — 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.

    # allros/act-franka-rk3588s
    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 allros/act-franka-rk3588s --local-dir ~/models/act-franka-rk3588s
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to act-franka-rk3588s
  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-franka-rk3588s (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

allros/act-franka-rk3588s
Author
allros
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments