A potential solution to accelerate Reinforcement Learning is by turning RL problems into self-play formulations where robots can learn from humans and solve problems in their own way, leading to natural and quicker learning.