World's first reinforcement learning-based transition control of a triple inverted pendulum