Release 0.83.5: Overmind-RL Reinforcement Learning Research

We’re pleased to announce version 0.83.5 of the Screeps GPT autonomous bot.

What’s New

New Features

  • Overmind-RL Reinforcement Learning Research: Comprehensive evaluation of RL integration potential for bot AI optimization
    • Created research documentation in docs/research/overmind-rl-analysis.md
    • Analyzed Overmind-RL three-component architecture (Node.js backend, Python Gym wrapper, distributed training)
    • Evaluated RL algorithms (PPO, DQN), neural network designs, and reward function engineering
    • Assessed 7 use cases: combat micro, resource allocation, expansion, creep bodies, market trading, tasks, pathfinding
    • Documented training requirements: $2k-$10k first year, 870 hours effort, GPU infrastructure
    • Detailed cost-benefit analysis: RL 6x more expensive than proven Overmind patterns with uncertain ROI
    • Compatibility analysis: Architecture misalignment (Python vs. TypeScript-only), 10-200ms inference latency
    • Created 6-phase integration roadmap (33-48 weeks) if pursued in future
    • Decision: NOT RECOMMENDED for current integration—focus on proven optimization patterns instead
    • Defined revisit conditions: bot maturity (12-24 months), specific high-value use case, RL expertise, infrastructure budget
    • Updated TASKS.md with research findings and alternative recommendations
    • Related research: Overmind architecture (overmind-analysis.md), creep-tasks (#625), packrat (#626)

Full Changelog: 0.83.5 on GitHub