9 summarised stories about Game AI, each linking back to the original source. Browse all topics →
OpenAI Blog 8 years ago
DeepMind trained AI agents through self-play competition to discover athletic skills including tackling, ducking, and catching without explicit instruction. The approach automatically maintains optimal difficulty as agents improve, similar to mechanisms observed in their Dota 2 experiments. This suggests self-play will become a standard component in developing capable AI systems.
OpenAI Blog 8 years ago
Meta-learning agents trained on simulated robot wrestling tasks can quickly adapt to defeat stronger opponents that lack meta-learning capabilities. The meta-learning approach enables agents to adjust their strategy during matches, including adapting when experiencing simulated physical malfunctions. This demonstrates that meta-learning could be useful for robotic systems that need to handle unexpected changes in their environment or body during operation.
