An Agile and Dexterous Balancing Mobile Manipulator Robot

Abstract

This thesis focuses on designing and controlling a dynamically stable shape-accelerating dual-arm mobile manipulator, the Carnegie Mellon University (CMU) ballbot. The CMU ballbot is a human-sized dynamically stable mobile robot that balances on a single spherical wheel. We describe the development of a pair of seven-degree-of-freedom (DOF) humanoid arms. The new 7-DOF arm pair have human-like kinematics, a large bimanual workspace, and the strength to carry a 20 kg payload. As part of this thesis work, the pair of strong and lightweight arms are integrated into the CMU ballbot. To the best of our knowledge, this robot configuration is the first and only of its kind.

The ballbot class of robots is inherently unstable and requires careful coordination between the upper and lower body to maintain balance while performing manipulation tasks. This thesis also demonstrates that this new configuration for mobile manipulation can be controllable over a wide envelope of possible configurations. Two different control strategies are presented: (i) a decoupled lower and upper body control strategy where the existing balancing controller compensates for the arm movement while the arms react to the body motion; and (ii) an optimal whole-body planning and control strategy that considers the entire kinematics and dynamics of the system in a single formulation.

The CMU ballbot already has an existing robust balancing controller. This controller was designed for the ballbot in its initial configuration without arms. It also assumes the center of mass (COM) is aligned with the central axis of the ballbot's cylindrical body. To balance the robot, the controller tracks a zero body lean angle. With the addition of the new arms, the COM will constantly move off-axis, and the zero body lean angle will not stabilize the system. Here we explore the control of the entire system's COM by controlling the body lean angle. This strategy decouples the balancing task from the upper body motions. We present joint space and task space controllers to control the new arms. This decoupled control strategy is evaluated through experiments on the CMU ballbot.

Most solutions to dynamic whole-body motion planning and control either use a complex, full-body nonlinear dynamic model of the robot or a highly simplified robot model. Here we explore centroidal dynamics, which has recently become a popular approach for designing balancing controllers for humanoid robots. We describe a framework where we first solve a trajectory optimization problem offline and later use the same nonlinear program (NLP) with a shorter time horizon in a model predictive control (MPC) context to execute the motion. We define balancing for a ballbot in terms of the centroidal momentum instead of other approaches like zero moment point (ZMP) or angular velocity that are more commonly used. We demonstrate that this framework can generate combined loco-manipulation motion plans and control inputs for the CMU ballbot.