- Command line running for headless terminals
Version 1.0 (Released November 2019)
Notes: The goal is to release this early Q4 2019 allowing the site to be made public
- Output to MP4 ✔️
- Output to PNG ✔️
- Support Basic Authentication ✔️
- Support API Keys in headers ✔️
- revise javadoc / unit tests ✔️
- build snapshots and upload code stats to website ✔️
Version 1.1 (Released November 2019)
- Change runtime statistics to better show the split between graphics / network / simulation processing. ✔️
- Show how long the simulation has been running. ✔️
- New thread management system to make scenario development easier. ✔️
- Twisty Puzzles ✔️
- Multi-Armed Bandit✔️
The following is a list of potential scenarios, and where they fit in the roadmap
- Car Driving (TBD) – Advanced car driving simulation including navigation and obstacle avoidance.
- Car Racing (Likely to be v1.5) – Top down racing game where the AI must drive round a track with simple physics.
- Nethack clone (TBD)
- Platform arcade game (TBD)
- Cart-Pole (v1.3) – Traditional physics simulation, balance a pole, keeping it upright for as long as possible.
- Snake (TDB)
- Sokoban Puzzle
- “Space Invaders” (TBD) – Clone of a 80’s style arcade game.
- Wargame (TBD)
- Zebra Puzzle (
v1.2) – A puzzle involving understanding the relationships between elements. Normally solved by constraints based programming, but could be adapted for learning algorithms. Zebra Puzzle page on Wikipedia. Note: This has been dropped from the 1.2 release.
The following would make great scenarios, but involve intellectual property / licencing issues; perhaps once we’re established we could approach the owners.
- Dracula’s dance (not sure of the name) – this is a print and play game that was previously available as part of a kickstarter campaign.
- Scotland Yard – board game where multiple agents have to work together to trap the ‘bad guy’ as they race across London
- Dominion – Deck building game created by Donald X. Vaccarino and distributed by Rio Grand. Would make for an interesting environment but potentially hard to code due to the volume of rules to be implemented. Also a lot of graphics assets to licence / replace.
- Hanabi – This game has been identified by several groups as a good target for AI research. Several online versions exist so it may be possible to implement the game however the publishers would still need to be approached.