Arena 3.0: Advancing Social Navigation in Collaborative and Highly Dynamic Environments

TL;DR

Arena 3.0 tackles the challenge of reliable social navigation in highly dynamic human environments by providing a cross-simulator platform that integrates diverse social force models, dynamic world generation, and an extended task and evaluation toolkit. The approach unifies three simulators (Flatland, Gazebo, Unity) under Arena-core with MBF-based navigation, an interplanner, and a Pedsim-based crowd model pipeline, enabling benchmarking across varied scenarios. Key contributions include the integration of multiple SFMs (including PySocialForce and Moussaid), a Pedsim Waypoint Plugin API for extensibility, social states and semantic mapping, and a rich task/world generation framework tested via user studies and large-scale benchmarks. The platform's cross-simulator abstraction, realistic crowd dynamics, and comprehensive metrics reduce sim2real gaps and support scalable research in social navigation and human-robot interaction.

Abstract

Building upon our previous contributions, this paper introduces Arena 3.0, an extension of Arena-Bench, Arena 1.0, and Arena 2.0. Arena 3.0 is a comprehensive software stack containing multiple modules and simulation environments focusing on the development, simulation, and benchmarking of social navigation approaches in collaborative environments. We significantly enhance the realism of human behavior simulation by incorporating a diverse array of new social force models and interaction patterns, encompassing both human-human and human-robot dynamics. The platform provides a comprehensive set of new task modes, designed for extensive benchmarking and testing and is capable of generating realistic and human-centric environments dynamically, catering to a broad spectrum of social navigation scenarios. In addition, the platform's functionalities have been abstracted across three widely used simulators, each tailored for specific training and testing purposes. The platform's efficacy has been validated through an extensive benchmark and user evaluations of the platform by a global community of researchers and students, which noted the substantial improvement compared to previous versions and expressed interests to utilize the platform for future research and development. Arena 3.0 is openly available at https://github.com/Arena-Rosnav.

Figures (6)

• Figure 1: Sample scenes from the Arena 3.0 platform, which provides tools to develop social navigation approaches in highly dynamic and crowded environments. It focuses on social navigation and provides a number of modules to achieve realistic simulation of human-centric environments, developing and testing navigation algorithms on various robotic systems and setups, and simplified extension with new modules.
• Figure 2: System architecture and modules of Arena 3.0. At its core, the platform consists of map and world generator algorithms, realistic pedestrian simulation, and comprehensive robot and navigation algorithm suites. The core architecture is fully abstracted from the simulators, allowing for cross-simulator scenario compatibility, with specific functionalities like RGB-D data available in Unity or Gazebo, and LiDAR applicable in all three simulators. Additionally, the platform features a training pipeline and an extended version of the MBF navigation framework to support the development and refinement of navigation approaches. Supplementary modules, such as the evaluation class and the web-based companion app, provide tools for data analysis and manual scenario creation. Using the provided API endpoints, the user can extend the platform with new planners (Move Base Flex API), task modes (task factory), social force models (SFM API), or simulators (simulator-factory). Further, computer vision modules for pedestrian detection and tracking are integrated to provide pedestrian data for further use.
• Figure 3: Differences between Arena versions: Arena 1.0 kastner2021arena (white), Arena 2.0 kastner2023arena (blue), and Arena 3.0 (green). Key enhancements include the complete abstraction of the Arena core, substantially extended map and world generation algorithms and task modes, additional robot and planners, provision of APIs for extensions, and a focus on social navigation, notably multiple realistic crowd behavior models, human-human and human-robot interactions, and social metrics.
• Figure 4: System design of the Pedsim Waypoint Plugin. Using this API enables users to seamlessly integrate new social force models and pedestrian behaviors, facilitating the inclusion of complex calculations and behaviors.
• Figure 5: Different map and world generator modes on three simulators. The map and world generation is abstracted from the simulators making the creation of worlds and scenarios unified across all simulators. This provides the ability to train and test the same agents on different simulators leveraging the strength of each simulator.