North Pole Madness using Deep Reinforcement Learning
A penguin must feed a baby. A bear must eat. And they play all by themselves. Short game implemented in Unity3D using the ML Agents library for deep reinforcement training. The penguin is rewarded when he picks up a fish and when it reach the baby.
Trained on CPU. Making multiple copies of the game arena can greatly speed up the process, allow for parallel training (suggested >8 arenas). Activate heuristic mode to manually control the penguin agent.
Initial positions randomly initialized.
- Penguin
- Pick a fish
- Avoid the bear
- Feed the baby
- Bear
- EAT THE PENGUIN !!!
Game finishes when there are no fish remaining in the pool or when the bear eat the penguin.
Unity3D 2019.4.18
ML-Agents 1.7.2
Barracuda 1.3.1
For any inquiries please contact: Alessandro Delmonte @ [email protected]
This project is licensed under the Apache License 2.0 - see the LICENSE file for details