README.md

metadata

env_name: Reacher-v5
tags:
  - Reacher-v5
  - ppo
  - reinforcement-learning
  - custom-implementation
  - mujoco
  - ddp
model-index:
  - name: PPO-DDP-ReacherV5
    results:
      - task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: Reacher-v5
          type: Reacher-v5
        metrics:
          - type: mean_reward
            value: '-5.30 +/- 1.20'
            name: mean_reward
            verified: false

PPO Agent playing Reacher-v5

This is a trained model of a PPO agent playing Reacher-v5.

Usage

create the conda env in https://github.com/GeneHit/drl_practice

conda create -n drl python=3.12
conda activate drl
python -m pip install -r requirements.txt

play with full model

# load the full model
model = load_from_hub(repo_id="winkin119/PPO-DDP-ReacherV5", filename="full_model.pt")

# Create the environment. 
env = gym.make("Reacher-v5")
state, _ = env.reset()
action = model.action(state)
...

There is also a state dict version of the model, you can check the corresponding definition in the repo.