A Synthetic Crowd Generation Framework for Socially Aware Robot Navigation

Socially aware robot navigation has gathered more and more interest from research communities due to its promising applications. Recent breakthroughs in Deep Reinforcement Learning (DRL) have opened many approaches to archive this task. However, due to the data-hungry characteristic of DRL-based methods, many promising proposed works have only trained on simulation, making real life applications still an open question. In this paper, we propose (i) a new Synthetic Crowd Generation (SCG) framework along with (ii) a world model for generating valid synthetic data. As a data-generating framework, SCG can be easily integrated into existing DRL-based navigation models without changing it. According to evaluations on simulation as well as real life data, our SCG has successfully boosted the published state-of-the-art navigation policy in terms of sample efficiency. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.