Abstract
This paper describes a modified sampling strategy for the Dynamic Window Approach (DWA), a local path planner, for omni-directional robot motion planning. An efficient local path planner allows the robot to quickly respond to dynamic obstacles and ensures that the resultant velocity commands meet the dynamic constraints of the robot. While typicalDWA implementations sample the velocity space evenly, we propose that a targeted sampling approach will result in a more fine-grained search of the relevant velocity space, leading to finer control and better performance in space-constrained environments. Our targeted sampling strategy (TS-DWA) is informed by the global planned path of the robot, allowing us to sample more velocities in the general path direction. We employ a polar velocity generator to selectively sample velocities and couple angular velocity samples to the path curvature. A bias for angular velocity is added for robots with a preferred heading, such as robots with forward-mounted sensors, to quickly turn towards the desired direction for better sensing. The strategy is implemented as a ROS Navigation Stack local_planner plugin and tested in simulation with Gazebo using an omni-directional robot platform. Experiments show that as the space around the simulated robot gets smaller, our proposed sampling strategy results in more successful navigation trials in space-constrained environments to the desired goal compared to other commonly-used methods like DWA and Timed-Elastic-Band, where planning fails or oscillates.