A Deep Learning Approach To Multi-Context Socially-Aware Navigation

TL;DR

This paper introduces a deep learning-based context classification system enabling robots to adapt their navigation strategies to different social scenarios, improving safety and comfort in human-robot interactions.

Contribution

It presents a novel high-level decision-making framework combining context classification with a multi-objective local planner for socially-aware navigation.

Findings

The robot successfully classified social contexts in real-time.

Navigation trajectories were socially appropriate and safe.

The system improved robot interaction quality in diverse environments.

Abstract

We present a context classification pipeline to allow a robot to change its navigation strategy based on the observed social scenario. Socially-Aware Navigation considers social behavior in order to improve navigation around people. Most of the existing research uses different techniques to incorporate social norms into robot path planning for a single context. Methods that work for hallway behavior might not work for approaching people, and so on. We developed a high-level decision-making subsystem, a model-based context classifier, and a multi-objective optimization-based local planner to achieve socially-aware trajectories for autonomously sensed contexts. Using a context classification system, the robot can select social objectives that are later used by Pareto Concavity Elimination Transformation (PaCcET) based local planner to generate safe, comfortable, and socially-appropriate trajectories for its environment. This was tested and validated in multiple environments on a Pioneer mobile robot platform; results show that the robot was able to select and account for social objectives related to navigation autonomously.