Inertial parameter estimation for robotic loaders from visual odometry and wheel-torques computed from control signals

Abstract

This thesis focuses on the study, validation, and comparison of methods for estimating inertial parameters of a robot based on physical variables obtained from various sources, including torque sensors, encoders, control signals, and visual information. Building upon previous work, specifically the stage of parameter calculation after estimation and filtering of variables used by different estimation methods, the research conducted experimental investigations using a modified semi-autonomous skid-steer loader. The loader was equipped with ROS (Robot Operating System) and auxiliary electronic boards, enabling the collection of sensor data and remote control of the machine. To obtain visual information, two Vicon Vantage V5 cameras were installed on the machine chassis, accompanied by software that communicated with Vicon Tracker. This software transformed motion data perceived by the cameras, which were designed for marker-based motion capture, into machine movement information. The resulting experiments provided data for analysis and evaluation of the proposed estimation methods. In general, this research contributes to the field of automation and control in heavy machinery by investigating the feasibility and effectiveness of using sensor data fusion techniques to estimate and track time-varying inertial parameters, enabling improved performance and operational capabilities in realworld applications.