Inference of Multi-Class STL Specifications for Multi-Label Human-Robot Encounters

This paper is interested in formalizing human trajectories in human-robot encounters. Inspired by robot navigation tasks in human-crowded environments, we consider the case where a human and a robot walk towards each other, and where humans have to avoid colliding with the incoming robot. Further, humans may describe different be-haviors, ranging from being in a hurry/minimizing completion time to maximizing safety. We propose a decision tree-based algorithm to extract STL formulae from multi-label data. Our inference algorithm learns STL specifications from data containing multiple classes, where instances can be labelled by one or many classes. We base our evaluation on a dataset of trajectories collected through an online study reproducing human-robot encounters.

I. Introduction

Social navigation of robots among humans is a research topic concentrating much interest due to diverse challenges, such as the prediction of human behavior in crowded social environments to enable efficient robot navigation, or the identification of “navigation styles” for a mobile robot in its surrounding environment [1]. In robot motion planning and control, desired behavior specifications and preferences can be expressed in Signal Temporal Logic (STL) [2], a formalism that can express system properties that include bounds on time and values of system parameters. STL defines specifications over multi -dimensional continuous signals and can, e.g., be used to depict properties of trajectories. Since STL allows rich expressiveness and resemblance to natural language [3], it has recently been used in human-robot interaction applications [4], [5], and human-robot encounters. Indeed, there is a growing need to depict human-robot in-teractions through formal and rigorous techniques to provide guarantees on the functioning of robotic systems, as well as their interactions with humans and perceived safety [6].