Meta-reinforcement learning capability for learning to explore

AI Impact Summary

The change introduces a capability to learn exploration strategies through meta-reinforcement learning. This shifts exploration policy development from hand-tuned heuristics to learned, task-adaptive strategies, affecting how RL training loops and data collection are designed. Engineering teams will need to extend pipelines to support meta-training, cross-task evaluation, and benchmarking of exploration performance. The business payoff is the potential for improved sample efficiency and generalization in RL workloads, at the cost of increased compute and integration effort.