CE-Nav is a learning-based, generalizable local navigation framework for robots. It features a novel two-stage (Imitation Learning-then-Reinforcement Learning) methodology that systematically decouples universal geometric reasoning from embodiment-specific dynamic adaptation, enabling efficient policy transfer across diverse robot morphologies including quadrupeds, bipeds, and quadrotors.

Go2	MagicDog	Spot

Hummingbird	H1

Completed:

• Cross-Embodiment Evaluation Framework - Unified evaluation methods for different robot platforms
• VelFlow Expert Model Checkpoint - Pre-trained General Expert model
• Go2 Model Checkpoint - Trained policy checkpoint for Unitree Go2 quadruped

Upcoming Releases:

• VelFlow Expert Training Code - General Expert model training pipeline
• Go2 Training Code - Complete training scripts for Unitree Go2 quadruped

Stay tuned for updates!

• NVIDIA GPU with CUDA support
• Ubuntu 20.04 / 22.04
• Conda

This project requires Isaac Sim version 2023.1.0-hotfix.1. Please ensure you download this exact version.

Step 1: Follow the Docker Container Setup.

Step 2: Download Isaac Sim to your docker container:

docker pull nvcr.io/nvidia/isaac-sim:2023.1.0-hotfix.1

docker run --name isaac-sim --entrypoint bash -it --runtime=nvidia --gpus all -e "ACCEPT_EULA=Y" --rm --network=host \
    -e "PRIVACY_CONSENT=Y" \
    -v ~/docker/isaac-sim/cache/kit:/isaac-sim/kit/cache:rw \
    -v ~/docker/isaac-sim/cache/ov:/root/.cache/ov:rw \
    -v ~/docker/isaac-sim/cache/pip:/root/.cache/pip:rw \
    -v ~/docker/isaac-sim/cache/glcache:/root/.cache/nvidia/GLCache:rw \
    -v ~/docker/isaac-sim/cache/computecache:/root/.nv/ComputeCache:rw \
    -v ~/docker/isaac-sim/logs:/root/.nvidia-omniverse/logs:rw \
    -v ~/docker/isaac-sim/data:/root/.local/share/ov/data:rw \
    -v ~/docker/isaac-sim/documents:/root/Documents:rw \
    nvcr.io/nvidia/isaac-sim:2023.1.0-hotfix.1

Step 3: Move the downloaded Isaac Sim from the docker container to your local machine:

docker ps  # Check your container ID in another terminal

# Replace <id_container> with the output from the previous command
docker cp <id_container>:/isaac-sim /path/to/local/folder  # Use absolute path

Clone the repository and set up the environment:

# Set the ISAACSIM_PATH environment variable
echo 'export ISAACSIM_PATH="/path/to/isaac_sim-2023.1.0-hotfix.1"' >> ~/.bashrc
source ~/.bashrc

# Navigate to the isaac-training directory
cd CE-Nav/isaac-training

# Run the setup script
bash setup.sh

After the setup completes, you should have created a virtual environment named cenav.

# Activate the environment
conda activate cenav

# Run evaluation to verify installation
cd training/scripts
python eval.py

If the installation is correct, you should see the Isaac Sim window open with the Go2 robot and obstacles.

Modify isaac-training/training/cfg/eval.yaml to configure the evaluation mode:

Dynamics-Aware Refiner Policy (Recommended):

evaluation:
  mode: guided_student
  expert_cnfm_model_path: "../../../il_training/fastsys/checkpoints/dynfji91/best_model.pt"
  student_checkpoint: "ckpts/checkpoint_7500.pt"

General Expert (VelFlow) Model:

evaluation:
  mode: expert_cnfm
  expert_cnfm_model_path: "../../../il_training/fastsys/checkpoints/dynfji91/best_model.pt"

Run Evaluation:

conda activate cenav
cd isaac-training/training/scripts
python eval.py

If you use CE-Nav in your research, please cite our paper:

@article{yang2025cenav,
  title={{CE-Nav: Flow-Guided Reinforcement Refinement for Cross-Embodiment Local Navigation}},
  author={Yang, Kai and Zhang, Tianlin and Wang, Zhengbo and Chu, Zedong and Wu, Xiaolong and Cai, Yang and Xu, Mu},
  journal={arXiv preprint arXiv:2509.23203},
  year={2025},
  eprint={2509.23203},
  archivePrefix={arXiv},
  primaryClass={cs.RO}
}

This project is licensed under the MIT License - see the LICENSE file for details.

This project builds upon several excellent frameworks. We are particularly grateful to:

• NavRL - The Isaac Sim training component of our framework is built upon NavRL
• Isaac Sim - NVIDIA's robotics simulation platform
• Orbit - Robot learning framework
• OmniDrones - Drone simulation framework
• TorchRL - PyTorch reinforcement learning library
• Unitree Robotics - Go2 quadruped robot