Vilas Boas, Lorenna Santos; https://orcid.org/0009-0009-9271-3852; https://lattes.cnpq.br/4253294933051903
Resumo:
Wheeled skid-steer robots are widely used in robotic applications of various types. Due to
non-holonomic constraints, a common characteristic of this class is wheel slippage during
rotational maneuvers, which hinders odometry. This dissertation investigates the kinematic
modeling applied to wheeled skid-steer mobile robots, based on the Instantaneous Center
of Rotation (ICR) and its effects on odometry. Aspects such as wheel slip, compensation
parameters for mechanical inaccuracies, and asymmetry between the left and right ICRs of
the robot are considered in the modeling. Two methods for ICR identification will be applied:
(1) the offline method, for identifying the ICRs through experiments involving rotation
and translation of the robot, or using a Genetic Algorithm that minimizes odometric
errors in simulated trajectories; and (2) the online method, applying the Extended Kalman
Filter (EKF) to estimate the location of the ICRs during the robot’s navigation, in order
to understand the effects of speed variation and changes in terrain characteristics on
the modeling. The UGV Husky A200 robot from Clearpath will be used in this study,
with absolute localization provided by optical motion capture via Optitrack. Laboratory
tests were conducted following a lemniscate trajectory at different speeds on dry and wet
industrial epoxy flooring. Results obtained from the ICR-based kinematic models show
a reduction in odometry pose errors by approximately 50 times compared to the ideal
differential model. Furthermore, EKF tests show that variations in wheel speed and floor
conditions impact the location of the ICR points, which is an important consideration for
modeling purposes. Thus, the contributions of this work highlight the improvements in
kinematic modeling with ICR for the wheeled skid-steer class and provide a powerful tool
for future applications in navigation algorithms for robots of this type.