Evaluation on IMU and odometry sensor fusion simulation results for a 4-Wheeled-Drive robot using AMCL on ROS framework.

Resumo

Mobile robot localization algorithms establish correspondence between the robot’s coordinate system
and the environment map. However, usually, it is not possible to get an accurate direct measurement of the robot
pose. Therefore, it must be inferred from local sensors, which are, in general, susceptible to noise. Some robot
models, such as the 4-Wheeled-Drive, can rapidly increase odometry error, especially due to drift during angular
movements. Hence, correction techniques are recommended to compensate these errors. This paper evaluates the
impact of the Extended Kalman Filter (EKF) to fuse data from an Inertial Measurement Unit (IMU) sensor and
odometry and how it impacts in the Adaptive Monte Carlo Localization (AMCL) algorithm. The use of data fusion
brings better results than using data from only encoders. The experiments were simulated using ROS framework
and Gazebo as simulation environment. A series of goals were given to test the localization algorithms performance
using the move base control algorithm in a map with obstacles. This experimental setup allowed to verify how
each localization method behaves comparing (a) only odometry; (b) odometry and the AMCL; (c) the fusion
between odometry and IMU, and (d) the fusion of odometry with AMCL and IMU.