Skip to content
Computer vision

microduck-skill-agent

LoRA adapters for robot control in MuJoCo

2 min readupdated

A set of LoRA adapters (lightweight add-ons for a base model) that turns the multimodal Qwen3-VL-4B-Instruct model into an action planner for a small walking "microduck" robot in the MuJoCo simulator. At each step, the model takes a camera frame and a text instruction in English or Chinese, then outputs a specific JSON command (such as move, turn, approach, explore, or return_home). Rather than acting as a direct motor controller, it handles only high-level skill selection and works alongside algorithms from the project repository (depth mapping and A* pathfinding). It is built for robotics researchers and developers exploring VLM (vision-language model) navigation in virtual environments.

Key features

  • 11 selectable skills: predicts a single action from move, turn, approach, search, explore, return_home, stop, done, clarify, unsupported, or bypass_return.
  • Bilingual commands: understands instructions in English and Chinese grounded in the live camera feed.
  • Four versions in one repository: includes lang-v1–lang-v4 subfolders, with lang-v2 serving as the main release with the strongest results.
  • Lightweight fine-tuned add-on: packs just 655,360 trainable parameters (LoRA rank 8) on top of a 4-bit quantized Qwen3-VL-4B-Instruct base.

What you can do

  • Test VLM navigation in MuJoCo: run the agent in multi-room simulated worlds to locate objects and reach targets (for example, "walk to the leftmost ball and return").
  • Compare training iterations: load different subfolders (lang-v1–lang-v4) to see how dataset scale impacts skill-selection accuracy.
  • Study hybrid architectures: use the project as a reference for pairing a language model with classical algorithms like A* and frontier-based map exploration.

Example

import torch
from transformers import AutoModelForImageTextToText, BitsAndBytesConfig
from peft import PeftModel

base = "Qwen/Qwen3-VL-4B-Instruct"
quant = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
                           bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16)
model = AutoModelForImageTextToText.from_pretrained(base, quantization_config=quant, device_map="auto")
model = PeftModel.from_pretrained(model, "atlasmarkz/microduck-skill-agent", subfolder="lang-v2")
Read moreDetails
lang-v2 benchmarks on test layouts (seeds 99000, 99100): strict accuracy 61/74 (82.4%), outcome-equivalent accuracy 65/74 (87.8%).
Training: 300 steps on 3,074 examples (for lang-v2) took roughly 13 minutes on a single RTX 3060 (8 GiB).
Limitations: intended strictly for simulation (do not run on physical hardware); requires a specific prompt format and the skill executors from the atlasmarkz/microduck-qwen repository.
License: Apache-2.0.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.3 GB · ready files9.6 GB
  • FP8 / INT8recommendedweights 4.1 GB5.0 GB
  • 4-bitweights 2.3 GB3.1 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 3080

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

This is a LoRA: counted together with its base model Qwen3-VL-4B-Instruct.

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.

    # atlasmarkz/microduck-skill-agent
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 Qwen/Qwen3-VL-4B-Instruct --local-dir ~/models/Qwen3-VL-4B-Instruct
    hf download atlasmarkz/microduck-skill-agent --local-dir ~/models/microduck-skill-agent
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3-VL-4B-Instruct
  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

Requires the base model

Qwen/Qwen3-VL-4B-Instruct

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of Qwen3-VL-4B-Instruct (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 5All on Hugging Face

About the model

atlasmarkz/microduck-skill-agent
Author
atlasmarkz
Category
Computer vision
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