Responsibility and Engagement -- Evaluating Interactions in Social Robot Navigation

TL;DR

This paper model the conflict buildup phase by introducing a time normalization and proposes the related Engagement metric, which captures how the agents'actions intensify a conflict, and shows that the metrics carry meaningful information about the cooperative resolution of conflicts in interactions.

Abstract

In Social Robot Navigation (SRN), the availability of meaningful metrics is crucial for evaluating trajectories from human-robot interactions. In the SRN context, such interactions often relate to resolving conflicts between two or more agents. Correspondingly, the shares to which agents contribute to the resolution of such conflicts are important. This paper builds on recent work, which proposed a Responsibility metric capturing such shares. We extend this framework in two directions: First, we model the conflict buildup phase by introducing a time normalization. Second, we propose the related Engagement metric, which captures how the agents' actions intensify a conflict. In a comprehensive series of simulated scenarios with dyadic, group and crowd interactions, we show that the metrics carry meaningful information about the cooperative resolution of conflicts in interactions. They can be used to assess behavior quality and foresightedness. We extensively discuss applicability, design choices and limitations of the proposed metrics.

Figures (9)

• Figure 1: Construction of the predicted Distance of Closest Encounter ($\textrm{pDCE}$). Based on their relative position and velocity, the time to closest encounter (TCE) can be computed. The distance at that time is the $\textrm{pDCE}$.
• Figure 2: Exemplary depiction of the relationship between Potential Conflict, Normalization, and resulting Conflict for a generic interaction scenario where two agents are on a collision course (c.f. Experiment 1, Oncoming, Scenarios 2 and 3). The Conflict Potential is high from the beginning, but the normalized Conflict starts building up only from $t=\text{TCE}-N_W$ onward. At $t>t_i$, one agent gradually changes its course to avoid the collision, causing the Conflict Potential to decrease. Accordingly, the Conflict buildup slows down, and, eventually, the Conflict also starts to decrease, until both Conflict Potential and Conflict reach zero shortly before $t=\text{TCE}$.
• Figure 3: Experiment 1 - Oncoming scenarios; Trajectories and Responsibility/ Engagement metrics for all scenarios. Arrow sizes indicate velocities, sampled at equidistant time intervals.
• Figure 4: Experiment 1 - Crossing scenarios
• Figure 5: Experiment 1 - Overtaking scenarios