Legged Ball Catching

This project is based on legged_gym and provides a framework for training, simulating, and deploying reinforcement learning policies for legged robots.

Directory Structure

The project adopts a modular design, clearly separating training, simulation, and deployment functions for easier development and maintenance:

legged_ball_catching/
├── legged_gym/    # Reinforcement learning framework based on NVIDIA Isaac Gym
├── mujoco_test/   # MuJoCo simulation environment and test scripts
│ ├── data/        # Robot model and environment configuration files
│ ├── test_script/ # Simulation test scripts
│ └── model/       # Trained policy model
├── deployment/    # Real robot deployment code
├── rsl_rl/        # Reinforcement learning algorithm implementation
├── README.md      # Project documentation
├── lcm_daemon.sh  # LCM daemon startup script
└── run_policy.sh  # Policy execution script

---

Environment Setup

This project involves three main stages: training, MuJoCo simulation, and deployment.
Please follow the recommended environment setup for each stage to ensure compatibility and reproducibility.

---

1. Clone the Repository

git clone https://github.com/Aiden2keke/legged_ball_catching.git
cd legged_ball_catching

2. Training Environment

The training code is based on legged_gym. Please follow the installation part in legged_gym docs for details to set up the environment and dependencies.

---

3. MuJoCo Simulation Environment

For MuJoCo-based simulation and visualization, install MuJoCo and its Python bindings in your environment:

pip install mujoco==3.2.2 mujoco-python-viewer

You may also need:

pip install pynput matplotlib

Note:

• Use a Python environment compatible with MuJoCo (Python 3.8+ recommended).
• For more details, refer to MuJoCo documentation.

---

3. Deployment Environment

For deployment on real robots, please follow the environment setup and deployment instructions from
walk-these-ways-go2.

Key steps:

• Build and install the deployment code as described in the above repository.
• Ensure all hardware and software dependencies are met for your robot platform.

---

Summary:

• Training: Follow legged_gym for environment and code base.
• MuJoCo Simulation: Install MuJoCo and related Python packages as above.
• Deployment: Follow walk-these-ways-go2 for environment and deployment steps.

If you encounter issues, please check the documentation of each upstream project or open an issue in this repository.

Training

1. Teacher-Student Training

Train a policy using the teacher-student framework:

python legged_gym/scripts/train.py --task=go2 --headless

2. Student Reinforcing Training

Continue training or reinforce a student policy:

python legged_gym/scripts/train.py --task=go2 --headless --max_iterations=1500 --student_reinforcing --resume --experiment_name=rough_go2 --run_name={run_name} --checkpoint=7500

3. Policy Evaluation (Play)

Evaluate a trained policy:

python legged_gym/scripts/play.py --task=go2 --load_run={run_name} --checkpoint=9000

---

MuJoCo Simulation

To run simulation and visualization in MuJoCo, use the provided scripts in mujoco_test/test_script/.
For example:

python mujoco_test/test_script/ball_catching.py

You can also run the simulation with the baseline policy as well.

---

Deployment

Follow walk-these-ways-go2 for the deployment steps.

First run lcm_position_go2 in the background:

sudo ~/program/legged_ball_catching/deployment/go2_gym_deploy/build/lcm_position_go2 eth0

Run this before running deploy_policy.py

cd deployment
export PYTHONPATH=$PYTHONPATH:`pwd`
cd go2_gym_deploy/scripts
python3 deploy_policy.py

---

Additional Notes

• For custom tasks or environments, refer to the legged_gym/envs and mujoco_test/test_script directories.
• For troubleshooting or further customization, consult the official legged_gym documentation and MuJoCo documentation.
• If you encounter issues with dependencies, ensure your Python environment matches the recommended versions.

---

Contact

For questions or contributions, please open an issue or pull request on the GitHub repository.