Autonomous mobile robot indoor navigation using multi-sensor integration

Description

Currently, most autonomous mobile robot indoor navigation systems are unable to provide absolute state information (e.g., coordinates in a reference frame) and rely on expensive sensors. The goal of this research is to develop a low-cost, high-accuracy, autonomous mobile robot indoor navigation system. The robot starts from an unknown location in a corridor environment and arrives at a selected target point with certain accuracy by following the centre line or virtually any lane of the corridors. The core research of this autonomous navigation system is in the development of reliable indoor orientation and position estimation algorithms. Integrating MEMS inertial and magnetic sensors improves overall performance of orientation estimation. However, challenges exist in dealing with the large gyro sensor errors and the large measurement noises of the accelerometers and magnetometers. A quaternion-based Kalman filter has been developed, which applies tightly-coupled and closed-loop integration strategies. It incorporates an online sensor calibration procedure for modelling time-varying sensor biases of the accelerometers and magnetometers, and a mechanism for adapting the measurement noise in the presence of motion and magnetic disturbances. In static mode, the integration algorithm can provide an estimation accuracy of less than o 1 when there is no magnetic anomaly. Even with the existence of significant magnetic disturbances, the orientation estimation error is reduced from up to o 131.6 to o 4.7 . In kinematic mode, the solutions show as much as 40% error reduction compared to those without applying the integration strategy. A novel indoor positioning system based on radio frequency identification technology has been developed, which can deal with complicated indoor radio signal environments due to multipath, non-line-of-sight, and signal interference. A regularized particle filter has been built by employing a non-parametric, probabilistic observation model. An effective online measurement quality control algorithm has been developed, which can identify and reject non-line-of-sight and/or multipath corrupted measurements. The developed indoor positioning system achieved a mean positioning error of 1.64 m, which is about 49% or more improvement in accuracy compared to other conventional methods. To successfully guide a robot to a target position, a sonic-vision system that can profile the local environment has been developed and two intelligent controllers have been designed. An efficient autonomous navigation algorithm has been developed, which choreographs all sub-system components comprising the orientation estimation module, the positioning module, the sonic-vision, and the intelligent controllers. The results showed that the robot is able to autonomously navigate to a pre-specified target point with a mean offset of 2.38 m. The average cross-track error was about 0.1 m which indicates the controllers’ autonomous capability in tracking and guidance. Overall results have confirmed the significant performance improvements of the developed orientation and position estimation methods, the benefits of applying them for indoor navigation, and the effectiveness of the autonomous navigation algorithm.