Domain Randomization Strategies for Sim-to-Real Transfer
Diagnose which gap—visual, physics, sensor, or scenario—then apply targeted randomization.
Svetlana Řehořová
Svetlana Řehořová
6 stories
Transformer Architectures as Policy Networks in Game Domains
Self-attention lets game AIs link distant board positions without recurrent bottlenecks.
Svetlana Řehořová
Exploration Strategies for Procedurally Generated Game Environments
Different algorithms encode different spatial grammars that guide what explorers should search for.
Svetlana Řehořová
Offline RL vs Online RL Tradeoffs for Fixed Game Datasets
Offline RL requires tight data coverage; online RL tolerates gaps by learning from mistakes.
Svetlana Řehořová
Hierarchical RL for Long-Horizon Strategy Game Tasks
Layering policies shortens planning horizons and splits credit assignment in long-horizon games.
Svetlana Řehořová
Multi-Agent Game Environments for Emergent Behavior Research
Competitive pressure alone produces complex strategies agents never learn to code.
Svetlana Řehořová