Robot Localization in Indoor and Outdoor Environments by Multi-sensor Fusion

Robot Localization (robot pose determination) has become an important aspect for a variety of tasks accomplished by mobile robots. Also accurate localization is required for robot tracking, path planning and control. Today, many sensor technologies are utilized to determine the exact robot location, for instance, in indoor environments, odometers (wheel encoders) and inertial navigation system (INS) are used to ascertain the relative position and pose of the robot. In outdoor environments, Global Positioning System (GPS) can also be integrated in the sensor suite to determine the actual position of the robot in terms of latitude and longitude. This paper presents a robust methodology for robot localization in indoor as well as outdoor environments by a mechanism known as "data fusion" of multiple sensors also called Multi Sensor Fusion (MSF) or Information Fusion (IF). In, outdoor environments, the sensor information collected from the embedded INS and GPS modules on mobile test robot are fused using a recursive state estimation and fusion algorithm known as Kalman Filter (KF). The estimated position obtained using KF is then combined with odometer based position data using weighting scheme to obtain the final position estimate of the robot. The main contribution of this work is to employ a multi-layer perceptron neural network (MLP-NN) to provide robot position estimates in an indoor environment where GPS signals are blocked. The MLP-NN is trained when the GPS data is available. As soon as the GPS signals are lost the trained MLP-NN provides predictions regarding the current position of robot. The proposed scheme is tested on the GPS-INS data obtained from on board sensors attached to a mobile robot. Simulation results have been presented which establish the efficacy of the proposed scheme.