This work presents planning and learning algorithms for average-cost MDPs with dynamic risk measures.

Actor-free Q-learning in Continuous Action Spaces by Learning a "Wire-fitted Q-function"