Trust, Transparency, and Explainable Behavior

People trust robots more when they can tell what the robot is about to do. A predictable motion is the one an observer expects for a known goal, usually the most efficient one. A legible motion is shaped so the observer can quickly infer the goal, even if that means a small detour. Robots also signal intent with lights, sounds, projected paths, and explanations.

Motion Exaggeration Goal

Inferring a goal from motion

An observer watching a motion from $S$ to the current point $Q$ judges how likely each goal $G$ is by how efficient the motion so far would be for that goal, following Dragan and Srinivasa:

$$P(G \mid S \to Q) \propto \frac{\exp\!\big(-C(S \to Q) - C^*(Q \to G)\big)}{\exp\!\big(-C^*(S \to G)\big)}$$

where $C$ is path length and $C^*$ the shortest remaining length. A legible motion maximizes how fast this probability rises for the true goal.

What to notice

The straight, predictable motion heads between the two objects at first, so the observer stays near 50-50 until late. The legible motion bends away from the other object early, and the observer becomes confident much sooner, at the cost of a slightly longer path. Too much exaggeration looks odd and can even mislead. Legibility is one part of transparent behavior. Others include signaling intent, explaining decisions in plain words, and showing confidence honestly, so people neither over-trust nor under-trust the robot.