Socially aware robot navigation, merging human-robot interaction and autonomous navigation, has gained prominence in recent years due to the rise of service robots operating in human environments. Although this field is receiving considerable global research attention, it still faces significant challenges in human-behavior modeling, contextual adaptation, simulations, datasets, design principles, and benchmarking – issues that call for community-wide standards rather than individual efforts. As a first step toward this goal, the current Dagstuhl Seminar is dedicated to tackling some of the key challenges and advancing universal standards in metrics and benchmarks. This four-day seminar will bring together experts in the fields of motion planning and control of robots, autonomous vehicles, human-robot interaction, and social psychology who share the collective knowledge required to answer the complex challenges of socially aware robot navigation.

The goal of the seminar is to address the major obstacles in socially aware robot navigation in order to move the field forward. By bringing together a unique group of researchers, it provides an opportunity to collaboratively explore these issues in an interdisciplinary setting. The key focus of this seminar will be on the following challenges:

Human Modelling: Humans are often modelled as dynamic obstacles to ease the design of the system. This model is limited and does not often scale well to complex situations and contexts. More information, like gaze, posture, age, etc., is required to better predict the motion and intentions in a given interaction. Data-based approaches need to be explored to obtain a more realistic human model.

Design Considerations for Autonomous Navigation System: Designing a robotic system that achieves societal acceptance is a challenging task. A lot of factors need to be taken into account, like the robot’s shape and design, environmental adaptability, communication and negotiation methodologies, legibility cues, and cultural background. Some of these require more studies, and the experts in social psychology and HRI could help to design them.

Data and Evaluation: Evaluation of socially aware navigation is still an open question, and there are no universal metrics or benchmarks that capture the interactions well. Even the community is split between various types of evaluation. There is also a need for good robot-human interaction datasets that can be used to validate these metrics or evaluate different navigation algorithms. Furthermore, evaluation and data collection through simulators are limited, as most of the existing robotic simulators lack good human simulation.

Learning and Foundation Models: Numerous works in the field employ learning techniques to model socially acceptable robot motion, which have the potential to capture complex behavior patterns. Building on this, the community could advance toward foundation models that enhance socially aware navigation by interpreting social cues, predicting human responses, and adapting to complex interactions. This requires fine-tuning with specialized datasets, safety through constraint satisfaction, and validation in both simulations and real-world deployments.

By bringing together the experts from this interdisciplinary field, the seminar will serve as a venue to address the existing challenges, define research directions, and lay the groundwork for the next generation of socially competent autonomous robots.

Creative Commons BY 4.0

Phani Teja Singamaneni, Anaís Garrell Zulueta, Claudia Pérez D'Arpino, Harold Soh, and Rachid Alami