Social Navigation

A robot that takes the shortest collision-free path through a crowd will brush past people and cut in front of them, which feels rude and unsafe. Social navigation adds the space people expect around themselves, larger in front where they are walking, so the robot gives way the way a polite person would.

Planner Personal space in front Personal space at the sides

A personal space cost

Proxemics, the study of personal space, suggests people feel discomfort when someone enters about 0.5 to 1.2 m around them. A common model is an uneven Gaussian, longer in front of a walking person, along their heading $\theta$:

$$c(\mathbf{x}) = \exp\!\left(-\frac{d_\parallel^2}{2\sigma_\parallel^2} - \frac{d_\perp^2}{2\sigma_\perp^2}\right), \qquad \sigma_\parallel = \begin{cases} \sigma_{\text{front}} & d_\parallel > 0 \\ \sigma_{\text{side}} & d_\parallel \le 0 \end{cases}$$

where $d_\parallel$ and $d_\perp$ are the offsets along and across the person's heading. The robot adds this cost when choosing its motion.

What to notice

The shortest route cuts straight through gaps between people and passes right in front of them, so the intrusion count climbs. The social planner slows, waits, or passes behind people instead, taking a little longer. Tuning the front and side sizes changes how cautious the robot is. Real systems also predict where each person will walk next, often with learned models.