A new approach to evaluating legibility: Comparing legibility frameworks using framework-independent robot motion trajectories
Robots that share an environment with humans may communicate their intent using a variety of different channels. Movement is one of these channels and, particularly in manipulation tasks, intent communication via movement is called legibility. It alters a robot's trajectory to make it intent expressive. Here we propose a novel evaluation method that improves the data efficiency of collected experimental data when benchmarking approaches generating such legible behavior. The primary novelty of the proposed method is that it uses trajectories that were generated independently of the framework being tested. This makes evaluation easier, enables N-way comparisons between approaches, and allows easier comparison across papers. We demonstrate the efficiency of the new evaluation method by comparing 10 legibility frameworks in 2 scenarios. The paper, thus, provides readers with (1) a novel approach to investigate and/or benchmark legibility, (2) an overview of existing frameworks, (3) an evaluation of 10 legibility frameworks (from 6 papers), and (4) evidence that viewing angle and trajectory progression matter when users evaluate the legibility of a motion.