This project explores deep reinforcement learning (DRL) within Unity, focusing on training an agent to jump over obstacles. The agent receives positive rewards for successfully jumping over a block and negative rewards for touching the obstacle. With this project I am learning more about neural networks and how to develop a fully functional DRL solution to teach the agent optimal behavior through trial and error.
This project depends on my Neural Network from scratch project, which I just copied into Unity.
The project is still under development, and the agent does not yet behave perfectly. Further training and optimization are needed to improve its performance and stability.
- Unity Integration: The project is made entirely in Unity.
- Reward System: Positive rewards for successful jumps and negative penalties for hitting obstacles.
- Deep Reinforcement Learning: Utilizes DRL to enable the agent to learn autonomously through feedback from its environment.
Initial training is yielding mixed results, with the agent sometimes failing to jump over the obstacles.
- Clone the repository.
- Open the project in Unity.
- Run the game to watch the agent in action and adjust training settings as needed.