Arena 4.0: A Comprehensive ROS2 Development and Benchmarking Platform for Human-centric Navigation Using Generative-Model-based Environment Generation

arXiv 2024

Volodymyr Shcherbyna, Linh Kästner, Diego Diaz, Huu Giang Nguyen, Maximilian Ho-Kyoung Schreff, Tim Lenz, Jonas Kreutz, Ahmed Martban, Huajian Zeng, Harold Soh
Paper arXiv Code

Abstract

Building on the foundations of our previous work, this paper introduces Arena 4.0, a significant advancement over Arena 3.0, Arena-Bench, Arena 1.0, and Arena 2.0. Arena 4.0 offers three key novel contributions: (1) a generative-model-based world and scenario generation approach that utilizes large language models (LLMs) and diffusion models to dynamically generate complex, human-centric environments from text prompts or 2D floorplans, useful for the development and benchmarking of social navigation strategies; (2) a comprehensive 3D model database, extendable with additional 3D assets that are semantically linked and annotated for dynamic spawning and arrangement within 3D worlds; and (3) a complete migration to ROS 2, enabling compatibility with modern hardware and enhanced functionalities for improved navigation, usability, and easier deployment on real robots. We evaluated the platform's performance through a comprehensive user study, demonstrating significant improvements in usability and efficiency compared to previous versions. Arena 4.0 is openly available at https://github.com/Arena-Rosnav.

Arena-Gen

Go2 quadruped in a generated world Pedestrians in a generated world Pedestrians in a scenario Classroom world Hospital world Factory world

Arena-Unity

Hawker centre world

Benchmark

One-click process for benchmarking a competition of

state-of-the-art planners

Trail MPC Dragon CADRL CrowdNav-ROS DWA APPLR TEB SARL* ROSNavRL RLCA-ROS

on over 10 different robots

Dingo
Dingo
Care-O-bot 4
Care-O-bot 4
Ridgeback
Ridgeback
Jackal UGV
Jackal UGV
Robotino
Robotino
TIAGo
TIAGo
youBot
youBot
TurtleBot3 Burger
TurtleBot3 Burger
TurtleBot3 Waffle Pi
TurtleBot3 Waffle Pi

in many realistic worlds

Warehouse
Warehouse
Classroom
Classroom
Japanese Garden
Japanese Garden
Hospital
Hospital

under limitless test scenarios

Hawker centre Factory COM3 Small warehouse Large hospital

with automatic data recording and post-processing

Collisions per planner Path length per planner Time looking at pedestrians Time looked at by pedestrians Episode results Time per planner

Training

In today's dynamic environments, traditional planning methods often struggle to adapt, while Deep Reinforcement Learning (DRL) shows promise in obstacle avoidance but faces challenges in long-range navigation due to local minima.

That's where we come in. We recognize the importance of combining the strengths of both traditional planning methods and DRL-based approaches to create a robust and efficient navigation system. Our platform offers a comprehensive framework that seamlessly integrates these two paradigms, providing you with the tools and resources needed to train and test deep reinforcement learning algorithms alongside classic methods.

Flexibility

  • With our platform, you have the flexibility to choose from a variety of robot models and define your own Deep Neural Network (DNN) or select from predefined networks.
  • Training is simplified through a user-friendly interface and a single config file, making the process accessible to users of all skill levels.

Control

  • Our platform supports multi-processed rollout collection for training with debug mode, ensuring efficient training and evaluation of your navigation system.
  • You can also enable and modify a custom training curriculum to tailor the training process to your specific requirements.

Seamless Integration

  • Once trained, your agent can be seamlessly deployed onto your mobile robot, ready to navigate dynamic environments with confidence.
  • With optional data logging, you can track the performance of your navigation system over time, ensuring continuous improvement and optimization.

BibTeX

@misc{shcherbyna2024arena4,
  title={Arena 4.0: A Comprehensive ROS2 Development and Benchmarking Platform for Human-centric Navigation Using Generative-Model-based Environment Generation},
  author={Volodymyr Shcherbyna and Linh K{\"a}stner and Diego Diaz and Huu Giang Nguyen and Maximilian Ho-Kyoung Schreff and Tim Lenz and Jonas Kreutz and Ahmed Martban and Huajian Zeng and Harold Soh},
  year={2024},
  eprint={2409.12471},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2409.12471}
}