Learning by Research

Thesis Title: Modelling and simulating realistic pedestrian behaviour in a shared space with an autonomous vehicle

In the near future, autonomous vehicles will have to navigate in dense and dynamic urban environments like shared spaces. In such spaces, segregation between pedestrians and vehicles is minimized and users must negotiate their passage without explicit traffic rules. Pedestrians navigate following some social norms and they expect the autonomous vehicle to navigate safely, efficiently and in accordance with the social and urban conventions. To achieve this, a key element of AV navigation in shared spaces is understanding and anticipating pedestrian behaviors and their interactions. However, we do not know yet how pedestrians will behave, because it is still very uncommon for pedestrians to share their space with autonomous vehicles. Our research problem is the following: how to anticipate pedestrian behaviors in a space shared with an autonomous vehicle? This thesis is part of the ANR project HIANIC (Human Inspired Autonomous Navigation In Crowds). In this thesis, we study the behavior of pedestrians in a shared space with an autonomous vehicle by modeling and simulating realistic pedestrian behavior. Our approach integrates empirical observations and concepts from social science into an agent-based model and simulator for an application in robotics. At each step, the proposed model has been evaluated and validated through simulations of many scenarios and comparisons with real-world data. Our first contribution is an agent-based model for individual pedestrian behavior in a shared space context. The model takes into account pedestrians' perception, attention and personal space in order to simulate sparse crowds in open environments. Our second contribution is an agent-based model for pedestrian groups with four social relationships (couples, friends, families and work colleagues). The model simulates both the movement of social groups of pedestrians in various crowd contexts and the avoidance behavior of groups by other pedestrians. Our third contribution is an agent-based model for pedestrian interaction with an autonomous vehicle in a shared space. The model allows to represent heterogeneous, accurate and explainable pedestrian behaviors in several interaction situations with an autonomous vehicle. The model can be used to reproduce real-world scenes and predict pedestrian trajectories around an autonomous vehicle in real time. Our fourth contribution is the implementation of the model in order to propose the simulator SPACiSS, "Simulator for Pedestrians and an Autonomous Car in Shared Spaces". SPACiSS is open source and can simulate interactions between pedestrians and vehicles in different shared space scenarios. With the integration in the ROS framework, commonly used in robotics, SPACiSS is designed as an environment to test autonomous navigation systems. We have shown that agent-based modeling and simulation supports the successful integration of social science and robotics. This association is promising to address real-world scenarios.

Key Words: Agent-based social simulation, Modelling, Pedestrians, Shared space, Autonomous car, Crowd

Doctoral School: ED MSTII - Mathématiques, Sciences et technologies de l'information, Informatique
Research Laboratory:  Laboratoire d’Informatique de Grenoble (LIG – CNRS/Inria/Grenoble INP-UGA/UGA)
Thesis Supervision: Julie DUGDALE and Anne SPALANZANI

> To know all the 2022 thesis prizes