Socially-Aware Opinion-Based Navigation with Oval Limit Cycles

TL;DR

It is shown that by combining opinion dynamics with vortex fields (to reach a consensus) with vortex fields (to generate socially acceptable trajectories), the result outperforms the application of the two techniques in isolation.

Abstract

When humans move in a shared space, they choose navigation strategies that preserve their mutual safety. At the same time, each human seeks to minimise the number of modifications to her/his path. In order to achieve this result, humans use unwritten rules and reach a consensus on their decisions about the motion direction by exchanging non-verbal messages. They then implement their choice in a mutually acceptable way. Socially-aware navigation denotes a research effort aimed at replicating this logic inside robots. Existing results focus either on how robots can participate in negotiations with humans, or on how they can move in a socially acceptable way. We propose a holistic approach in which the two aspects are jointly considered. Specifically, we show that by combining opinion dynamics (to reach a consensus) with vortex fields (to generate socially acceptable trajectories), the result outperforms the application of the two techniques in isolation.

Figures (6)

• Figure 1: The general diagram of our framework.
• Figure 2: Passage preference, i.e., rotation direction $\gamma$, depending on the value of $z$: if $z < 0$, then $\gamma = 1$, i.e. the vortex field is clockwise; vice versa, if $z > 0$, then $\gamma = -1$, i.e. the vortex field is counterclockwise.
• Figure 3: Human-robot passing simulation: human and robot trajectories towards their targets (stars). The lines change from solid to dashed when the robot enters the limit cycle, i.e. when its attention level starts to rise.
• Figure 4: Comparison between approaches: red line for robot trajectory, blue line for human trajectory, red star for robot target.
• Figure 5: Simulation test with oval limit cycle for both agents. The lines change from solid to dashed when they enter the other's limit cycle.