Stochastic neural networks added to hierarchical reinforcement learning capability

AI Impact Summary

A new capability adds stochastic neural networks to the hierarchical reinforcement learning (HRL) toolkit, enabling stochastic policies and probabilistic sub-policy modeling. This can improve exploration efficiency and long-horizon policy performance, particularly in noisy or uncertain environments. Teams should review documentation for API changes, adapt training pipelines to support stochastic components, and plan experiments to quantify robustness and compute trade-offs.