Credenz'25 Xodia - Angry Birds RL Competition

Welcome to the Angry Birds RL Competition! 🎯🐦
This competition challenges participants to develop Reinforcement Learning (RL) agents that can efficiently launch birds to hit pigs using a physics-based environment.

📌 Game Overview

The environment simulates an Angry Birds-style game where:

• A bird is launched from a slingshot towards a target pig.
• The bird follows a trajectory influenced by gravity.
• Players control the launch power and angle to maximize accuracy.

---

🏗 Observation Space

The environment provides an observation space represented as a 8-dimensional vector:

Feature	Description	Range
bird_x	Bird's x-coordinate	[0, 800]
bird_y	Bird's y-coordinate	[0, 450]
velocity_x	Bird's x-axis velocity	[-30, 30]
velocity_y	Bird's y-axis velocity	[-30, 30]
pig_x	Pig's x-coordinate	[0, 800]
pig_y	Pig's y-coordinate	[0, 450]
max_height	Maximum height reached by the bird	[0, 450]
launch_angle	Bird's launch angle (radians)	[-π, π]

---

🎮 Action Space

Participants control the bird's launch power along the x and y axes:

Action	Description	Range
power_x	Launch power in x-direction	[1, 15]
power_y	Launch power in y-direction	[1, 15]

• A higher power_x pushes the bird farther horizontally.
• A higher power_y results in a higher arc.

---

🏆 Scoring & Rewards

Your RL agent earns rewards based on performance:

• ✅ +200 for hitting the pig.
• 🎯 +0.5 per unit decrease in distance to the pig.
• 🏹 Bonus reward for achieving interesting trajectories.
• ❌ -5 for going out of bounds.
• ⏳ Episode ends when:
  • The bird hits the pig.
  • The bird goes out of bounds.
  • The maximum number of steps (200) is reached.

---

🛠 Implementation Tasks

To successfully participate, complete the following key components:

1️⃣ Implement the Training Loop (train.py)

• Load the Angry Birds RL environment.
• Initialize the RL agent.
• Implement the training loop where the agent:
  • Selects actions based on its policy.
  • Interacts with the environment.
  • Collects rewards and updates its model.
• Save the trained model for evaluation.

2️⃣ Complete the Reward Function (AngryBirdsEnv)

• Ensure rewards follow the scoring system outlined above.
• Implement logic for:
  • Positive rewards for hitting the pig.
  • Penalty for going out of bounds.
  • Reward shaping to encourage better trajectories.

---

Good luck to all the participants!