Isaac-Gym-Jackal

Prerequisite

1. Isaac Gym 1.0.preview4
2. https://github.com/NVIDIA-Omniverse/IsaacGymEnvs

Environment Explanation

• To perceive obstacles, image information is given to RL agents.
• We preprocess the image information uisng VAE, and the encoded vectors are given as observations.
• Therefore, the environment is constructed as follows:
  1. Images are collected in vae/collect.py with the DummyJackal task.
  2. Encder and Decoder are trained in vae/main.py.
  3. Using the trained encoder, the Jackal task is constructed in utils/jackal_env/task2.py.

How to train RL agents

# 1. Collect images.
cd vae
python collect.py
# 2. Train VAE.
python main.py
# 3. Train PPO agents.
cd ../ppo
python main.py
# 4. Visualize PPO agents.
python main.py --test