Reinforcement learning framework adds variational option discovery algorithms

AI Impact Summary

New capability: variational option discovery algorithms are introduced in the reinforcement learning toolkit. These methods automatically learn temporally extended actions by optimizing a variational objective, improving exploration in long-horizon tasks. Teams should anticipate new APIs or parameters to enable option discovery, plus potential compute overhead from variational inference and additional hyperparameters to tune. This can accelerate development of hierarchical policies and unlock more scalable RL pipelines once properly validated.