Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/spartan-71/pocket-tanks
Reinforcement Learning Agent for the ultimate AI War (Credenz '24)
https://github.com/spartan-71/pocket-tanks
pocket-tanks reinforcement-learning-agent stable-baselines3
Last synced: 19 days ago
JSON representation
Reinforcement Learning Agent for the ultimate AI War (Credenz '24)
- Host: GitHub
- URL: https://github.com/spartan-71/pocket-tanks
- Owner: Spartan-71
- License: apache-2.0
- Created: 2024-04-17T17:32:43.000Z (9 months ago)
- Default Branch: main
- Last Pushed: 2024-12-29T19:41:19.000Z (22 days ago)
- Last Synced: 2024-12-29T20:18:50.035Z (22 days ago)
- Topics: pocket-tanks, reinforcement-learning-agent, stable-baselines3
- Language: Jupyter Notebook
- Homepage: https://pypi.org/project/Xodia24/
- Size: 20.5 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
XODIA Reinforcement Learning Competition
🏆 3rd Place Winner | April 2024 | Pocket Tanks AI Competition
## 🎮 About The Project
This repository showcases my solution for the **XODIA Reinforcement Learning Competition**, where I secured **3rd place** among 30+ participants. The challenge involved developing an AI agent capable of mastering *Pocket Tanks* through optimized reward function engineering and reinforcement learning techniques.
### 🎯 Competition Objectives
- Design an intelligent AI bot for Pocket Tanks
- Implement an optimized reward function
- Compete against other AI agents in various scenarios
- Maximize performance and strategic decision-making
## 🛠️ Technical Stack
### Core Technologies
```
🐍 Python 3.8+
🤖 Xodia24 (Competition Framework)
🧠 stable-baselines3
🔥 PyTorch
📊 TensorBoard (Monitoring & Visualization)
☁️ Google Colab (Training Environment)
```
## 🧮 Reward Function Architecture
Our sophisticated reward system employs advanced mathematical modeling to optimize agent behavior:
1. **Advanced Mathematics**
- Quadratic equations for precision control
- Linear decay patterns for predictable behavior
- Hyperbolic functions for specialized scenarios
2. **Distance-Based Scaling**
- Dynamic reward adjustment based on target distance
- Optimized range effectiveness calculations
- Strategic positioning incentives
3. **Seven Bullet Types**
- Standard Shell: Close combat specialist
- Triple Threat: Multi-range effectiveness
- Long Shot: Distance warfare
- Heavy Impact: Maximum damage potential
- Blast Radius: Area control
- Healing Halo: Support capabilities
- Boomerang Blast: Tactical specialty
4. **Strategic Design**
- Engineered for tactical diversity
- Balanced risk-reward mechanics
- Situation-aware decision making
## 🏆 Competition Results
### Achievements
- 🥉 **3rd Place** Overall Ranking
- 📈 Consistent High-Performance Metrics
- 🎯 Superior Strategic Decision Making
### Watch the Competition
▶️ [AI Wars Showcase](https://youtu.be/fUzpJypN_Hg?si=EIBE7uiDjvoIliQ_)
## 🙏 Acknowledgments
- The XODIA organizing team for creating this challenging competition
- Fellow participants for pushing the boundaries of AI gaming
- The reinforcement learning community for valuable resources
## 📬 Contact
For questions or collaboration opportunities, feel free to reach out!
---
Made with 🤖 and ❤️ for the XODIA Competition