README.md

metadata

license: cc-by-nc-sa-4.0
tags:
  - robotics
  - vision-language-action-model
  - vision-language-model
pipeline_tag: robotics
library_name: transformers

Model Card for InternVLA-M1_object

InternVLA-M1 is an open-source, end-to-end vision–language–action (VLA) framework for building and researching generalist robot policies, as introduced in the paper: InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy.

• 🌐 Homepage: InternVLA-M1 Project Page
• 💻 Codebase: InternVLA-M1 GitHub Repo

Abstract

We introduce InternVLA-M1, a unified framework for spatial grounding and robot control that advances instruction-following robots toward scalable, general-purpose intelligence. Its core idea is spatially guided vision-language-action training, where spatial grounding serves as the critical link between instructions and robot actions. InternVLA-M1 employs a two-stage pipeline: (i) spatial grounding pre-training on over 2.3M spatial reasoning data to determine where to act'' by aligning instructions with visual, embodiment-agnostic positions, and (ii) spatially guided action post-training to decide how to act'' by generating embodiment-aware actions through plug-and-play spatial prompting. This spatially guided training recipe yields consistent gains: InternVLA-M1 outperforms its variant without spatial guidance by +14.6% on SimplerEnv Google Robot, +17% on WidowX, and +4.3% on LIBERO Franka, while demonstrating stronger spatial reasoning capability in box, point, and trace prediction. To further scale instruction following, we built a simulation engine to collect 244K generalizable pick-and-place episodes, enabling a 6.2% average improvement across 200 tasks and 3K+ objects. In real-world clustered pick-and-place, InternVLA-M1 improved by 7.3%, and with synthetic co-training, achieved +20.6% on unseen objects and novel configurations. Moreover, in long-horizon reasoning-intensive scenarios, it surpassed existing works by over 10%. These results highlight spatially guided training as a unifying principle for scalable and resilient generalist robots.

🔥 Key Features

1. Modular & Extensible
   All core components (model architecture, training data, training strategies, evaluation pipeline) are fully decoupled, enabling independent development, debugging, and extension of each module.

2. Dual-System and Dual-Supervision InternVLA-M1 integrates both a language head and an action head under a unified framework, enabling collaborative training with dual supervision.

3. Efficient Training & Fast Convergence Learns spatial and visual priors from large-scale multimodal pretraining and transfers them via spatial prompt fine-tuning. Achieves strong performance (e.g., SOTA-level convergence on in ~2.5 epochs without separate action pretraining).

🎯 Target Audience

1. Users who want to leverage open-source VLMs (e.g., Qwen2.5-VL) for robot control.
2. Teams co-training action datasets jointly with multimodal (vision–language) data.
3. Researchers exploring alternative VLA architectures and training strategies.

📊 Experimental Results

	WindowX	Google Robot(VA)	Google Robot(VM)	LIBERO
$\pi_0$	27.1	54.8	58.8	94.2
GR00t	61.9	44.5	35.2	93.9
InternVLA-M1	71.7	76.0	80.7	95.9

🚀 Quick Start

🛠 Environment Setup

# Clone the repo
git clone https://github.com/InternRobotics/InternVLA-M1

# Create conda environment
conda create -n internvla-m1 python=3.10 -y
conda activate internvla-m1

# Install requirements
pip install -r requirements.txt

# Install FlashAttention2
pip install flash-attn --no-build-isolation

# Install InternVLA-M1
pip install -e .

⚡ Quick Interactive M1 Demo

Below are two collapsible examples: InternVLA-M1 chat and action prediction.

InternVLA-M1 Chat Demo (image Q&A / Spatial Grounding)

from InternVLA.model.framework.M1 import InternVLA_M1
from PIL import Image
import requests
from io import BytesIO
import torch

def load_image_from_url(url: str) -> Image.Image:
    resp = requests.get(url, timeout=15)
    resp.raise_for_status()
    img = Image.open(BytesIO(resp.content)).convert("RGB")
    return img

saved_model_path = "/PATH/checkpoints/steps_50000_pytorch_model.pt"
internVLA_M1 = InternVLA_M1.from_pretrained(saved_model_path)

# Use the raw image link for direct download
image_url = "https://raw.githubusercontent.com/InternRobotics/InternVLA-M1/InternVLA-M1/assets/table.jpeg"
image = load_image_from_url(image_url)
question = "Give the bounding box for the apple."
response = internVLA_M1.chat_with_M1(image, question)
print(response)

InternVLA-M1 Action Prediction Demo (two views)

from InternVLA.model.framework.M1 import InternVLA_M1
from PIL import Image
import requests
from io import BytesIO
import torch

def load_image_from_url(url: str) -> Image.Image:
    resp = requests.get(url, timeout=15)
    resp.raise_for_status()
    img = Image.open(BytesIO(resp.content)).convert("RGB")
    return img

saved_model_path = "/PATH/checkpoints/steps_50000_pytorch_model.pt"
internVLA_M1 = InternVLA_M1.from_pretrained(saved_model_path)

image_url = "https://raw.githubusercontent.com/InternRobotics/InternVLA-M1/InternVLA-M1/assets/table.jpeg"
view1 = load_image_from_url(image_url)
view2 = view1.copy()

# Construct input: batch size = 1, two views
batch_images = [[view1, view2]]  # List[List[PIL.Image]]
instructions = ["Pick up the apple and place it on the plate."]

if torch.cuda.is_available():
    internVLA_M1 = internVLA_M1.to("cuda")

pred = internVLA_M1.predict_action(
    batch_images=batch_images,
    instructions=instructions,
    cfg_scale=1.5,
    use_ddim=True,
    num_ddim_steps=10,
)
normalized_actions = pred["normalized_actions"]  # [B, T, action_dim]
print(normalized_actions.shape, type(normalized_actions))

Training Details

action_chunk: 8
batch_size: 128
training_steps: 30k

For more detailed training scripts and datasets, please refer to the InternVLA-M1 GitHub Repo.

📈 Model Zoo

We release a series of pretrained models and checkpoints to facilitate reproduction and downstream use.

✅ Available Checkpoints

Model	Description	Link
InternVLA-M1	Main pretrained model	🤗 Hugging Face
InternVLA-M1-Pretrain-RT-1-Bridge	Pretraining on RT-1 Bridge data	🤗 Hugging Face
InternVLA-M1-LIBERO-Long	Fine-tuned on LIBERO Long-horizon tasks	🤗 Hugging Face
InternVLA-M1-LIBERO-Goal	Fine-tuned on LIBERO Goal-conditioned tasks	🤗 Hugging Face
InternVLA-M1-LIBERO-Spatial	Fine-tuned on LIBERO Spatial reasoning tasks	🤗 Hugging Face
InternVLA-M1-LIBERO-Object	Fine-tuned on LIBERO Object-centric tasks	🤗 Hugging Face

🤝 Contributing

We welcome contributions via Pull Requests or Issues. Please include detailed logs and reproduction steps when reporting bugs.

📜 Citation

If you find this useful in your research, please consider citing:

@article{internvlam1,
  title   = {InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy},
  author  = {InternVLA-M1 Contributors},
  journal = {arXiv preprint arXiv:2510.13778},
  year    = {2025}
}

📬 Contact

• Issues: Submit via GitHub Issues with detailed logs and steps

🙏 Acknowledgements

We thank the open-source community for their inspiring work. This project builds upon and is inspired by the following projects (alphabetical order):

• IPEC-COMMUNITY: Curated OXE / LIBERO style multi-task datasets and formatting examples.
• Isaac-GR00T: Standardized action data loader (GR00T-LeRobot).
• Qwen2.5-VL: Multimodal input/output format, data loader, and pretrained VLM backbone.
• CogACT: Reference for a DiT-style action head design.
• Llavavla: Baseline code structure and engineering design references.
• GenManip Simulation Platform: Simulation platform for generalizable pick-and-place based on Isaac Sim.

---

Thanks for using InternVLA-M1! 🌟 If you find it useful, please consider giving us a ⭐ on GitHub.