torobo_genesis

This repository is forked from Genesis (Commit ID:98b29b4a4e56bdaa55ca1d8a359c5935eb24f26c) and includes additions for reinforcement learning with hand_v4 model and leg_v1 model.

Installation

1. Setting up a Virtual Environment

   Python 3.9 or higher is required. Create a virtual environment using the following commands:

   conda create -n genesis python=3.9
   conda activate genesis

2. Installing PyTorch

   Install PyTorch. Refer to the official website below and execute the command suitable for your environment.

   PyTorch Official Website

   (Example: For CUDA 12.2)

   pip install torch torchvision torchaudio

3. Cloning and Installing torobo_genesis

   Clone this repository and install the required dependencies:

   git clone https://github.com/TokyoRobotics/torobo_genesis.git
   cd torobo_genesis
   pip install -e .

4. Installing rsl_rl

   Install rsl_rl, which is required for reinforcement learning:

   git clone https://github.com/leggedrobotics/rsl_rl
   cd rsl_rl && git checkout v1.0.2 && pip install -e .

   (Optional) To install TensorBoard:

   pip install tensorboard

Examples

Command Examples

Run reinforcement learning for hand_v4 model:

python examples/manipulation/hand_v4_train.py -v --num_envs 1024

-v: Visualize the training process.
--num_envs: Specify the number of environments. The default is 4096.

Run inference using the trained model:

python examples/manipulation/hand_v4_eval.py --ckpt 1000

--ckpt: Specify the checkpoint of the trained model.

To monitor the training process, launch TensorBoard using the command below:

tensorboard --logdir logs

Task Examples

Reorientation Task by hand_v4 model:

python examples/manipulation/hand_v4_train.py -v

python examples/manipulation/hand_v4_eval.py

Locomotion Task by leg_v1 model:

python examples/locomotion/leg_v1/leg_v1_train.py -v

python examples/locomotion/leg_v1/leg_v1_eval.py -v

License and Acknowledgments

The torobo_genesis source code is licensed under Apache 2.0 and BSD 3-Clause License.

Genesis's development has been made possible thanks to these open-source projects:

• Taichi: High-performance cross-platform compute backend. Kudos to the Taichi team for their technical support!
• FluidLab: Reference MPM solver implementation.
• SPH_Taichi: Reference SPH solver implementation.
• Ten Minute Physics and PBF3D: Reference PBD solver implementations.
• MuJoCo: Reference for rigid body dynamics.
• libccd: Reference for collision detection.
• PyRender: Rasterization-based renderer.
• LuisaCompute and LuisaRender: Ray-tracing DSL.