Using Unsupervised Learning to Explore Robot-Pedestrian Interactions in Urban Environments

TL;DR

This paper tackles the lack of data-driven, contextual understanding of robot-pedestrian interactions in urban environments by proposing an unsupervised learning pipeline. The approach fuses static maps and dynamic mission data, extracting context and interaction features, then reduces dimensionality with PCA before applying K-means clustering to reveal interaction patterns. A use-case on the RoboTraces dataset demonstrates the feasibility of the pipeline and reveals that data-driven clusters can differ meaningfully from human-defined, junction-based segmentation, while also highlighting the need for richer contextual representations and larger datasets. The work offers practical insights for improving situational awareness and interaction quality in real-world robot deployments, guiding mission planning and control strategies in smart-city contexts.

Abstract

This study identifies a gap in data-driven approaches to robot-centric pedestrian interactions and proposes a corresponding pipeline. The pipeline utilizes unsupervised learning techniques to identify patterns in interaction data of urban environments, specifically focusing on conflict scenarios. Analyzed features include the robot's and pedestrian's speed and contextual parameters such as proximity to intersections. They are extracted and reduced in dimensionality using Principal Component Analysis (PCA). Finally, K-means clustering is employed to uncover underlying patterns in the interaction data. A use case application of the pipeline is presented, utilizing real-world robot mission data from a mid-sized German city. The results indicate the need for enriching interaction representations with contextual information to enable fine-grained analysis and reasoning. Nevertheless, they also highlight the need for expanding the data set and incorporating additional contextual factors to enhance the robots situational awareness and interaction quality.

Figures (7)

• Figure 1: An exemplary conflict scenario of the Robot Claudi on a sidewalk.
• Figure 2: Pipeline for data-driven, bottom-up analysis of human-robot interactions in urban delivery scenarios.
• Figure 3: Part of the robot's trajectory in Freiberg, it covers different parts of the city (city center, low-traffic side streets, park). For each point at the trajectory, a distance information to the next junction node is calculated ranging from green at $\qty{65}{\meter}$ to red at $\qty{0}{\meter}$ distance. Positions closer than $\qty{8}{\meter}$ to a junction are highlighted by black color.
• Figure 4: Movements of pedestrians and the robot in different spatial constellations: a parallel movement, a lateral encounter and a head-on confrontation. Black dots represent the robot's GNSS positions. Human movement is shown by green dots. In both cases the red dot denoting the latest human position. Gray lines illustrate the connection between the robot's observation points and the human's positions. $\alpha$ indicated the angle between the two movements. Ranging from 0 indicating parallel movement to $\pi$ head-on movement.
• Figure 5: Scatter plot of the minimum distance between robot and pedestrian and the orientation of the directions of movement ($\qty{0}{\radian}\leq\alpha\leq\qty[parse-numbers=false]{\pi}{\radian}$)