Main menu
AI vs Human
AI training visualized with reward system and difficulty levels
AI vs AI for continuous learning

Inspiration

Re_Knight was inspired by the AI-driven robotic enemies in ARC Raiders and the idea of enemies learning behavior instead of relying only on fixed scripts. We wanted to apply that concept to a fighting game.

What it does

Re_Knight is a 2D fighting game with reinforcement-learning AI opponents. Players can fight the AI, watch AI vs AI battles, and experiment with a mode where an AI learner improves through combat.

How we built it

We built the game in Godot and trained the AI using reinforcement learning. The AI uses the same basic actions as the player, including movement, jumping, crouching, dodging, and attacking. We trained different skills in stages before combining them into full combat.

Challenges we ran into

The hardest part was getting the AI to learn useful behavior instead of repeating ineffective strategies. We also had to integrate the trained model into Godot while keeping the game systems compatible with it.

Accomplishments that we're proud of

We successfully trained an AI fighter, exported the model, and made it playable inside the game. We also created AI vs Human, AI vs AI, and live combat statistics.

What we learned

We learned that reinforcement learning depends heavily on rewards, observations, training difficulty, and environment design. We also learned how to take an AI model from training into a real playable game.

What's next for Re_Knight

Next, we want to improve player adaptation, expand Rags to Riches, add more enemies and arenas, and polish the game with better sound, visuals, and effects.