A team of researchers that includes Claas Völcker and Igor Gilitschenski from the University of Toronto developed REPPO, a new AI training method that improves speed, reliability and efficiency for reinforcement learning systems used in robotics and autonomous technologies.