UnityRLAgent

Train a Simple AI Agent in Unity Using C# and Q-Learning

UnityRLAgent is a straightforward AI agent for Unity that uses Reinforcement Learning (Q-Learning) to learn how to reach a target. Built entirely with C#, it requires no external AI libraries, making it easy to integrate and visualize the learning process.

Key Features

• Simple AI Agent: Learn to navigate towards a stationary target using basic movements and rotations.
• Visual Learning Process: Watch the agent improve its behavior over time directly in the Unity scene.
• Configurable Learning Parameters: Easily adjust learning rate, discount factor, exploration rate, and more.
• No External Dependencies: Fully implemented in C# for seamless integration into your Unity projects.
• Easy to Extend: Simple project structure allows for easy customization and feature additions.

Usage

• Start Training: Click the Play button in Unity. The agent will begin moving towards the target.
• Adjust Parameters: Modify the agent's learning and movement parameters in the Inspector to see how they affect training.
• Observe Learning: Watch as the agent gradually improves its ability to reach the target more efficiently.

Configuration

Agent.cs Variables

• moveStep (float)

  • Description: Distance the agent moves with each action.
  • Impact: Higher values speed up movement but reduce precision. Smaller values make movement more accurate but slower.

• rotationStep (float)

  • Description: Degrees the agent rotates with each action.
  • Impact: Larger angles allow quicker direction changes. Smaller angles provide smoother directional adjustments.

• learningRate (float)

  • Description: How much new information overrides old information.
  • Impact: Higher values lead to faster learning but can cause instability. Lower values make learning more stable but slower.

• discountFactor (float)

  • Description: Importance of future rewards.
  • Impact: Higher values encourage long-term planning. Lower values focus more on immediate rewards.

• explorationRate (float)

  • Description: Probability of choosing a random action.
  • Impact: Higher values promote exploration of actions. Lower values make the agent rely more on known best actions.

• explorationDecay (float)

  • Description: Rate at which exploration decreases over time.
  • Impact: Slower decay keeps exploration active longer, allowing the agent to continue discovering new actions.

• minExplorationRate (float)

  • Description: Minimum exploration probability.
  • Impact: Ensures the agent always has some chance to explore, preventing it from getting stuck on suboptimal actions.

• maxStepsPerEpisode (int)

  • Description: Maximum steps per training episode.
  • Impact: Limits the duration of each episode to ensure consistent training and prevent endless attempts.

• stepDelay (float)

  • Description: Delay between each training step for visualization.
  • Impact: Set to 0 for faster training or increase to slow down and observe the agent's behavior.

Contributing

If you find this project useful and have ideas for improvements, feel free to open an issue or submit a pull request. Contributions are welcome!

For direct questions, reach out on Telegram: @CodeOrDie42.

License

This project is open-source and available under the MIT License.

---

UnityRLAgent helps you understand the basics of Reinforcement Learning in Unity, providing a foundation for building more complex AI systems. Happy coding! 🚀