Automated Playing of Survival Video Games with Commonsense Reasoning (short paper)
summary
An agent for Don't Starve converts in-game Lua JSON into predicate facts, applies about 22 rules and 8 assumptions encoded in answer set programming, and uses the s(CASP) goal-directed engine with a Lua mod bridge to select actions. Eight runs show survival strategies (gather, craft, cook, run) that reach up to day seven, dying mainly to darkness, starvation, or hostile mobs. The prototype outperforms novice play but lacks long-range memory and optimal resource planning.
AI-written summary · about
chapters (10)
- 0:00Automated survival game playing
- 1:25Common sense reasoning framework
- 3:11Don't Starve game mechanics
- 5:52Agent world data interpretation
- 8:47Rule-based action selection
- 10:36Agent survival strategies
- 10:58Live demonstration
- 15:33Experimental results
- 16:17Conclusion and future work
- 18:26Q&A
related talks
27mCan computers become intelligent through playing games?
19mPrepare yourself against Zombie epidemic
01h 01mIntroduction to Practical AI
52mThe Journey of AI in Games: A Technical and Philosophical overview
43mRoleplAI
59mGame AI 4N00bz
39mBuilding self evolving video game AI in Python
39mEmbarrassing Mistakes: a short story about AI