Human-robot physical interaction with dynamically stable mobile robots

• Authors: Umashankar Nagarajan;George Kantor;Ralph L. Hollis
• Venue: Proceedings of the 4th ACM/IEEE international conference on Human robot interaction
• Year: 2009

Developed by Prof. Ralph Hollis in the Microdynamic Systems Laboratory at Carnegie Mellon University, Ballbot is a dynamically stable mobile robot moving on a single spherical wheel providing omni-directional motion. Unlike statically stable mobile robots, dynamically stable mobile robots can be tall and skinny with high center of gravity and small base. The ball drive mechanism is a four motor inverse mouse-ball setup. An Inertial Measuring Unit (IMU) and encoders on the motors provide all information needed for full-state feedback. Ballbot has three legs that provide static stability when powered down and is capable of auto-transitioning from the statically stable state to the dynamically stable state and vice versa. It is also capable of yaw rotation about its vertical axis. An absolute encoder provides the relative angle between the IMU and the ball drive unit. We wish to demonstrate Human-Robot Physical Interaction with dynamically stable mobile robots using Ballbot as an example. The balancing controller on Ballbot is extremely robust to disturbances like shoves, kicks and collisions with furniture and wall. Due to its dynamic stability, Ballbot can be moved around with very little effort. Physically directing a heavy statically stable mobile robot can be a difficult task, whereas Ballbot can be moved around with just a single finger. Similarly, while moving, Ballbot can be stopped with very little effort. We have developed some basic behaviors that enable Ballbot to detect human intentions with the physical interaction it has using just the encoder and IMU data. For example, given a soft push, Ballbot tries to stick to its position on the floor, whereas, when given a hard push, it moves away from its current location and station-keeps at a different point on the floor. We also present our initial results in developing a Learn-Repeat behavior in Ballbot, where in during the Learn mode, the user drives Ballbot around and it remembers the path travelled, and during the Repeat mode, Ballbot attempts to repeat the path learnt. We are in the process of adding stereo cameras and laser range finders to the robot, which will help us explore and extend more areas of Human-Robot Interaction.