Motion Tracking of Four-Wheeled Mobile Robots in Outdoor Environments Using Bayes’ Filters

The demand for outdoor wheeled mobile robots (WMRs) is rapidly growing to assist humans in outdoor environments such as transportation, exploration, rescue, security, agriculture, military, etc. To effectively control outdoor WMRs, we need motion models applicable in such unknown environments. Conventional modeling mainly concerns physics laws or equations of motion, such as kinematics, dynamics, terrains, and wheel-ground interactions. Modeling WMRs on unstructured ground is more complicated than in a well-developed indoor environment. To alleviate such difficulties, we looked at a data-driven approach instead. We built a four-wheeled mobile robot with wheel encoders installed, with which the forward and inverse differential kinematic solutions were derived. Then, we performed more than a thousand test runs in outdoor environments, having the robot run on normal, icy, and sandy roads, including test runs under constraints partially blocked by brick and grid-type holes. We employed Bayes’ filter because the robot’s tri-variate states (two linear velocities and the rotation) are not directly measurable through wheel encoders. With such uncertainty, Bayes’ filtering technique of the Kalman filter and a newly developed unscented Kalman filter were applied to infer how each wheel’s speed affects the robot’s velocity. We established a probabilistic motion model, where the differential kinematic solutions are combined with uncertainty from outdoor environments. Consequently, we could closely track the robot’s motion. This modeling technique can be used to develop better outdoor WMRs.