Skip to content
Other

X-VLA + Recap v1

robot control policy for cable insertion

2 min readupdated

The X-VLA + Recap v1 checkpoint trains a robotic arm for precision cable insertion in the EEE 598 AI for Industry Challenge (AIC). Built on the xvla-aic-v4 fine-tune for the LeRobot library, this 0.88B-parameter model is further fine-tuned using Recap (advantage-conditioned reinforcement learning) across 120 rollout recordings and 585 demonstrations. It achieved the highest total benchmark score, outperforming the base fine-tune as well as DSRL and residual RL variants. Useful for researchers and developers working in robotics and reinforcement learning.

Key features

  • Scores 1163 points across 36 test trials in the AIC Gazebo simulator, beating the base xvla-aic-v4 fine-tune at 1052 points.
  • Takes feeds from three wrist-mounted cameras (256x288) alongside a 19-value state vector (TCP pose, velocities, wrist forces, and torques).
  • Fine-tuned for 20k steps on the aic-recap-v1 dataset using one-second segments labeled with Advantage: positive and Advantage: negative tags.
  • Runs in best-policy mode by appending the " Advantage: positive" suffix to the task prompt.

What you can do

  • Test the policy in the AIC Gazebo simulation environment on precision positioning and cable insertion tasks using the eee598-aic project scripts.
  • Use the checkpoint as a baseline policy or reference when benchmarking offline and advantage-conditioned fine-tuning methods in LeRobot.

Example

hf download muhammad-ibraheem/xvla-recap-v1 --local-dir ckpt/xvla-recap-v1
scripts/run_smolvla_server.sh ckpt/xvla-recap-v1
Read moreDetails
Benchmark (36 trials in AIC Gazebo, max 100 points per attempt): 1163 points total (368 on Seed 100, 348 on Seed 101, 447 on training boards) and 2 full insertions.
Comparison: beats X-VLA + DSRL (1077 points), base xvla-aic-v4 (1052 points), X-VLA + residual RL (1011 points), and the second iteration xvla-recap-v2 (949 points).
Training: 20k steps on 120 X-VLA rollouts (top 30% and successful insertions labeled as positive) and 585 CheatCode demonstrations.
Discuss Telegram 0 17

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.6 GB · ready files2.3 GB
  • FP8 / INT8recommendedweights 0.8 GB1.4 GB
  • 4-bitweights 0.5 GB1.0 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.

    # muhammad-ibraheem/xvla-recap-v1
    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 muhammad-ibraheem/xvla-recap-v1 --local-dir ~/models/xvla-recap-v1
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to xvla-recap-v1
  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 xvla-recap-v1 (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

muhammad-ibraheem/xvla-recap-v1
Author
muhammad-ibraheem
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

170

Ask AI about this model

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

Comments

Other

CAT

CAT