Yann LeCun is not optimistic about reinforcement learning: "I do prefer MPC"  

Yann LeCun, Meta's Chief AI Scientist, expressed a greater fondness for Model Predictive Control (MPC) as representing the magic of planning in solving new tasks without the need for task-specific learning. He advocates minimizing the use of reinforcement learning in favor of greater use of model prediction to learn world models and task goals. However, he also recognizes that solving exact MPC models is a challenge, and that reinforcement learning and MPC may each have their own applicability scenarios. (Heart of the Machine)