Reinforcement learning competition pushes the boundaries of embodied AI