*2.1. Sampling-Based and Velocity-Based Methods*

Autonomous navigation in dynamic environments is challenging, since the planner is forced to frequently adjust its planned results for handling dynamic objects such as moving obstacles. Velocity obstacle (VO) [5] can simply consider dynamic constraints to compute trajectories of robots via the concept of velocity obstacles that denote the robot's velocities causing a collision with

obstacles soon. However, computing maneuvers by velocity obstacles has low efficiency for real-time applications. Sampling-based methods such as PRM and RRT are popular in robotics, benefitting from advantages regarding efficiency and robustness [15]. This kind of method can solve high-dimensional planning problems by approximating C-space. Nevertheless, in dynamic environments, PRM needs to recheck edge connection [16], and RRT is required to modify the pre-built exploring tree [15], which further decrease time efficiency when environments become more dynamic. Consequently, the above-mentioned model-based methods still suffer from essential disadvantages that cannot realize time-saving navigation in dynamic environments, so as to hinder the robot's navigation capabilities.
