Researchers at OpenAI have developed robots that can perform backflips purely based on human feedback, demonstrating the potential of machine learning and reinforcement learning in robotics.