Real-Time Avoidance Strategy of Dynamic Obstacles via Half Model-Free Detection and Tracking With 2D Lidar for Mobile Robots

Abstract

Avoidance is a necessary capability for a mobile robot to perform tasks, such as delivering objects in household or industrial scenarios. The existing avoidance strategy based on timed elastic band local planner and cost-map provided by robotics operating system cannot realize the excellent performance when a robot and an obstacle both move. In this article, we present a real-time, simple, and reliable approach to detecting and tracking obstacles via a two-dimensional lidar in dynamic scenarios where the mobile robot and the obstacle are moving. Obstacles are represented by a set of points against their outlines and the information of obstacles is initialized and updated via the raw laser measurement. First, the obstacle is detected by three main steps: preprocessing, segmentation, and merging, classification of consequent measurements. Second, we use a hierarchical method to realize data associations for figuring out the corresponding matches among obstacles with the consecutive time. Last, after doing the data association, we need to estimate the motion of the dynamic obstacle for being tracked by the Kalman filter. Extensive experiments performed in the simulation and practical scenarios indicate that the proposed method enables a mobile robot to perform dynamic avoidances efficiently [Real-time Avoidance Strategy of Dynamic Obstacles via Half Model-free Detection and Tracking (T-MECH)].