I. Introduction
Mobile robots are increasingly integrated into human life [1], [2]. The robot receives sensor data and computes kinodynamic-constrained trajectories in real-time [3]. This motion planning process can be completed by velocity-based methods that sample candidate trajectories in velocity space, such as Dynamic Window Approach [4] and Reciprocal Velocity Obstacle [5], or state-based methods that fit trajectories by candidate goal states, such as lattice planner [6]. However, they suffer from either a short horizon or tracking error. As a result, an optimization-based Timed Elastic Band (TEB) [7], [8], [9] forms a time-optimal trajectory in a long horizon and guarantees safe separation from obstacles.