Extending Teach and Repeat to Pivoting Wheelchairs

摘要

The paper extends the teach-and-repeat paradigm that has been successful for the control of holonomic robots to nonholonomic wheelchairs which may undergo pivoting action over the course of their taught movement. Due to the nonholonomic nature of the vehicle kinematics, estimation is required -- in the example given herein, based upon video detection of wall-mounted cues -- both in the teaching and the tracking events. In order to accommodate motion that approaches pivoting action as well as motion that approaches straight-line action, the estimation equations of the Extended Kalman Filter and the control equations are formulated using two different definitions of a nontemporal independent variable. The paper motivates the need for pivoting action in real-life settings by reporting extensively on the abilities and limitations of estimation-based teach-and-repeat action where pivoting and near-pivoting action is disallowed. Following formulation of the equations in the near-pivot mode, the paper reports upon experiments where taught trajectories which entail a seamless mix of near-straight and near-pivot action are tracked.