Terrain sensing and estimation for dynamic outdoor mobile robots

Abstract

In many applications, mobile robots are required to travel on outdoor terrain at high speed. Compared to traditional low-speed, laboratory-based robots, outdoor scenarios pose increased perception and mobility challenges which must be considered to achieve high performance. Additionally, high-speed driving produces dynamic robot-terrain interactions which are normally negligible in low speed driving. This thesis presents algorithms for estimating wheel slip and detecting robot immobilization on outdoor terrain, and for estimating traversed terrain profile and classifying terrain type. Both sets of algorithms utilize common onboard sensors. Two methods are presented for robot immobilization detection. The first method utilizes a dynamic vehicle model to estimate robot velocity and explicitly estimate longitudinal wheel slip. The vehicle model utilizes a novel simplified tire traction/braking force model in addition to estimating external resistive disturbance forces acting on the robot. The dynamic model is combined with sensor measurements in an extended Kalman filter framework. A preliminary algorithm for adapting the tire model parameters is presented. The second, model-free method takes a signal recognition-based approach to analyze inertial measurements to detect robot immobilization. Both approaches are experimentally validated on a robotic platform traveling on a variety of outdoor terrains. Two detector fusion techniques are proposed and experimentally validated which combine multiple detectors to increase detection speed and accuracy. An algorithm is presented to classify outdoor terrain for high-speed mobile robots using a suspension mounted accelerometer. The algorithm utilizes a dynamic vehicle model to estimate the terrain profile and classifies the terrain based on spatial frequency components of the estimated profile. The classification algorithm is validated using experimental results collected with a commercial automobile driving in real-world conditions.