https://github.com/kickflip-games/three_body_game

ludum dare 50 (https://avivajpeyi.itch.io/three-body)
https://github.com/kickflip-games/three_body_game

Last synced: 6 days ago
JSON representation

ludum dare 50 (https://avivajpeyi.itch.io/three-body)

• Host: GitHub
• URL: https://github.com/kickflip-games/three_body_game
• Owner: kickflip-games
• Created: 2022-04-04T00:03:04.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2022-04-04T01:06:07.000Z (over 2 years ago)
• Last Synced: 2024-10-19T12:53:29.386Z (27 days ago)
• Language: ShaderLab
• Size: 1.45 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Three Body Game

* [itch](https://avivajpeyi.itch.io/three-body)

* [ludum-dare-50](https://ldjam.com/events/ludum-dare/50/three-body)

As gravity pulls the three bodies close, destruction is inevitable. Try to weave between the two celestial bodies that are hurtling towards you to score points and delay the inevitable. 

## To train ML model:

We are using reinforcement learning with immitation-learing to initialise the training. 

1. Start `ML training` scene

2. Open terminal and source ML-agents python env

3. cd to project dir and run: ` mlagents-learn Assets/ML-Agents/Configs/threeBody

.yaml --run-id `

4. To view training progress run:  `tensorboard --logdir results/`

Current bot is not performming too well -- maybe the rewards are not adjusted? Maybe I should train using immitation learning for the first 5k steps, then switch over to training with reinforement learning?