Towards Humanoids Using Personal Transporters: Learning to Ride a Segway from Humans

Human bipedal locomotion is efficient, robust and versatile, but typically reserved to reach targets in the close vicinity. As soon as larger distances have to be covered, humans tend to rely on wheeled modes of transport in the form of cars, bikes, scooters etc. Having the flexibility to choose a personal transporter (PT) such as a Segway when needed, is also an interesting option for humanoids operating in the real world, but it requires the ability to control a device that has its own complex dynamics. In this paper, we synthesize controllers for the the humanoid robot REEM-C to ride a Segway in simulation, motivated by human Segway riding. We perform motion capture experiments of a human riding a Segway and identify human whole-body behavior as well as the Segway's internal controllers. We then show that the REEM-C can successfully generate translational, rotational and mixed motions of the Segway in simulation. The Segway is controlled by targeted motions of the REEM-C using an inverted pendulum based LQR controller for pitch control and an admittance controller for the LeanSteer to command a yaw-rate. After these promising simulation results, the next step will be implementation on a real Segway and the REEM-C humanoid.