Joint Potential-Vector Fields for Obstacle-Aware Legible Motion Planning
Huy Quyen Ngo, Aaron M. Steinfeld · 2024
Traditionally, potential fields and vector fields have been extensively used for motion planning, especially in finding paths to a goal position while avoiding obstacles along the way. However, such methods have only been applied to the problem of finding the shortest path to only one goal position. In human-centered environments with multiple goals, the shortest path (e.g., most predictable) is often not the most intent-expressive path (e.g., most legible) to one of the goals. We devised a method for robot planning and navigation in human-centered environments that uses potential fields to plan intent-expressive motion to a specific goal among many, while utilizing adaptive vector fields to avoid obstacles without sacrificing the legibility property of the motion. We found that our method can produce motions that are of comparable legibility and shorter path length compared to a legible motion planner baseline, as well as more legible paths compared to a traditional potential field method. Our method was evaluated in several scenarios where legibility is useful, namely maps with and without obstacles and goal switching.