Design, implementation and evaluation of a motion control scheme for mobile platforms with high uncertainties

In this work, we present a motion control scheme for a robotic mobile platform using low-cost vision sensor to update encoder values. We track the pose of a power wheelchair using wheel encoders along with a Microsoft Kinect camera. Two methods of pose estimation are implemented and tested. These methods are a) encoder-based odometry and b)ICP(Iterative Closest Point)-based updated odometry. We evaluate the performance of each method using precise wheelchair pose ground truth data acquired via a state-of-the-art VICON® system with eight motion capture cameras. Offline data processing is performed to refine the ICP parameters and estimate the covariance matrices of the Kalman filter. The offline data processing results demonstrate that our ICP-based updated odometry has very accurate pose tracking. By implementing our control scheme, the position error is improved by a factor of 15 and the localization orientation error is improved by a factor of 13. In online implementation, there was 4 times improvement for both position and orientation angle estimation. To demonstrate the robustness of our approach, we apply it for online obstacle avoidance. A wheelchair-mounted robotic arm (WMRA) is also included in this platform and will be used for future work on combined mobility and manipulation control with sensor assistance.