Demonstrating Arena 5.0: A Photorealistic ROS2 Simulation Framework for Developing and Benchmarking Social Navigation

arXiv:2610.11220 · cs.RO · Submitted 2026-10-08 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "Demonstrating Arena 5.0".

Dev: The gist Arena 5.0 provides three main contributions: 1) The complete integration of NVIDIA Isaac Gym, enabling photorealistic simulations and more efficient training,

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: So, looking at the title and who wrote this, we see authors from National University of Singapore, Technical University Berlin, and Technical University Munich all contributing to this Arena five point zero framework <ref:2610.11220#pg1>.

Dev: It’s a big international effort, which suggests they’re trying to make something robust enough for diverse research groups to use across different platforms and setups.

Taro: I’m interested in what that means for the actual autonomy research community—it looks like they are providing a unified way to approach social navigation testing.

Rosa: Right. They aren't just proposing a new algorithm; they are giving us the entire simulation environment, which is usually where most of the heavy testing happens.

Dev: And that means researchers can focus less on getting the simulator to run and more on actually developing their navigation strategies within a realistic setting.

Taro: So, it’s about giving people a reliable sandbox to prove their methods work before they ever put them out in the field.

The paper's summary: Rosa: Now, diving into what Arena five point zero actually does, the authors say their main contributions are three things: first, integrating Isaac Gym for photorealistic simulations and faster training; second, creating a comprehensive benchmark for state-of-the-art social navigation strategies across many different worlds; and third, building these scenario generation tools for creating customized scenarios like emergency situations <ref:2610.11220#pg1>.

Dev: That means they’ve put together a whole system that lets you test your navigation plans against a wide variety of challenges, not just simple maze problems.

Taro: I see the focus on generating customized scenarios specifically for things like emergency and rescue situations; that sounds really practical for testing safety protocols.

Rosa: Yes, they also added enhanced sensor integration, supporting everything from 2D lidar to RGBD cameras and foot contact sensors, which helps cover a lot more ground in terms of what kind of navigation approaches you can test <ref:2610.11220#pg1>.

Dev: And the way they handle generation is pretty smart too; they use generative AI and Large Language Models to help create these environments intuitively through natural language prompts or by uploading floorplans.

The paper's improvements: Rosa: What’s really interesting about the improvements is how they’ve made the world creation process more accessible, using tools like 3dsg from text and the MiDiffusion network to turn text descriptions into finished three dee environments <ref:2610.11220#pg1>.

Dev: That’s a significant step because building those complex three dee scenes used to be a huge manual lift, but now it seems like you can describe what you want and get the environment built <ref:2610.11220#pg1>.

Taro: That ease of creation is important, but I wonder how well those generated environments actually behave when things go wrong; that’s where I get curious.

Rosa: Well, they say this system allows users to generate dynamic environments that include both static and dynamic obstacles, which means you can test how robots handle unpredictable movement in those complex settings.

Dev: And on the evaluation side, they aren't just looking at basic navigation accuracy; they’ve added social metrics like "Time seen by pedestrians s" and "Time in personal space s," which gives a much better picture of how good the human-robot interaction is actually being.

Conclusion: Rosa: So, to wrap up this paper on Demonstrating Arena five point zero, we’re seeing a framework that combines photorealistic simulation power with powerful generative tools for scenario building and detailed social metrics <ref:2610.11220#pg1>.

Dev: It really seems like they are providing researchers with a complete toolkit to move from simple testing to more complex, realistic evaluations of social navigation performance across diverse settings.

Taro: What this means for us is that we have a much better way to stress test our plans against unexpected human behavior and environmental complexity before we even think about deployment.

Rosa: Exactly. It’s about narrowing that gap between simulation and reality by making the simulation itself a more accurate reflection of the real-world interaction dynamics.

Dev: So, Arena five point zero gives us a much more capable platform for rigorous benchmarking of social navigation methods using those new, detailed metrics they introduced <ref:2610.11220#pg1>.

Taro: I think the focus on customization through scenario generation pipelines is key; it lets you tailor the test to exactly what part of social interaction you want to investigate next.

Rosa: That’s a great summary of it all, covering the integration, the generation tools, and the evaluation metrics in this paper on Demonstrating Arena five point zero <ref:2610.11220#pg1>.

National University of Singapore (NUS) · Technical University Berlin (TUB) · Technical University Munich (TUM)

cs.RO

Submitted: 2026-10-08

Updated: 2026-10-08

Comments: 12 pages, 10 figures. Published at Robotics: Science and Systems (RSS) 2025

Journal ref: Proceedings of Robotics: Science and Systems (RSS), Los Angeles, CA, USA, June 2025

DOI: 10.15607/RSS.2025.XXI.092

Code: https://github.com/ros-planning/navigation2

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 77/100

The gist: The gist Arena 5.0 provides three main contributions: 1) The complete integration of NVIDIA Isaac Gym, enabling photorealistic simulations and more efficient training, 2) A comprehensive benchmark of

Key concepts

NVIDIA Isaac Gym Integration
This involves fully incorporating the Isaac Gym simulation engine into the Arena platform. It allows researchers to create highly photorealistic simulations of robot interactions in complex environments, leading to more efficient training and testing of social navigation algorithms compared to older simulators.
Comprehensive Benchmarking Suite
The framework establishes rigorous tests for social navigation strategies. These benchmarks evaluate how accurately, efficiently, and in real-time robots can navigate human-centric simulated worlds using a wide range of metrics that cover accuracy and performance.
World and Scenario Generation Modules
These new modules use generative AI and Large Language Models (LLMs) to create social navigation environments intuitively. Users can generate diverse settings, such as offices or hospitals, either by uploading floor plans or by providing a natural language description of the desired scenario.

Terminology

Summary

The gist

Arena 5.0 provides three main contributions: 1) The complete integration of NVIDIA Isaac Gym, enabling photorealistic simulations and more efficient training, 2) A comprehensive benchmark of state-of-the-art social navigation strategies, evaluated on a diverse set of generated and customized worlds and scenarios of varying difficulty levels, and 3) Extensive scenario generation and task planning modules for improved and customizable generation of social navigation scenarios.

How it works

Arena 5.0 is built upon a modular software stack designed to support the development and benchmarking of social navigation methods (Page 1). This framework introduces three key contributions: (1) the integration of Isaac Gym, which allows for photorealistic simulations, (2) a comprehensive benchmarking suite for state-of-the-art social navigation methods incorporating multi-modal sensor data across various robotic platforms, and (3) novel world and scenario generation modules that employ generative AI and Large Language Models (LLMs) to facilitate the intuitive creation of social scenarios and environments (Page 3).

Isaac Gym Integration

The platform features a full integration of Isaac Gym, which includes a comprehensive ROS integration of Isaac Gym, facilitating seamless inter-operation with ROS-based systems (Page 2). This addition required several adaptations and new modules to fully leverage the capabilities offered by the Arena framework (Page 3). By embedding Isaac Gym into the Arena framework, researchers can effortlessly generate social navigation scenarios that exhibit photorealistic behaviors (Page 3). The platform leverages Isaac Gym’s advanced simulation capabilities, enabling the creation of complex interactive environments where robotic systems can be tested under conditions that closely mimic real-world settings (Page 3). This approach not only improves the utility of Isaac Gym but also expands its applicability to a broader range of social navigation research scenarios, thereby contributing to the advancement of human-robot interaction studies (Page 3).

World and Scenario Generation

Arena 5.0 offers distinct worlds where randomized environments can be generated across multiple simulation platforms including Gazebo and Isaac Sim (Page 4). These worlds encompass a variety of settings such as residential areas, hospitals, offices, and warehouses (Page 4). The algorithms employ a method of gradual generation, capable of creating dynamic environments that include both static and dynamic obstacles (Page 4). The user can either upload a 2D floorplan or input a natural language prompt that describes an environment (Page 4). This process involves several steps: first, the user prompts to either write an input prompt or upload a floor plan which is converted into a 3DSG and edited by the user (Page 4); second, a floorplan is generated by the HouseDiffusion and post-processed (Page 4); third, objects are placed into the rooms and realistically arranged by the MiDiffusion network, creating a world (Page 4).

Benchmarking and Evaluation

Arena 5.0 establishes rigorous benchmarks for evaluating the performance of social navigation algorithms, including metrics for assessing accuracy, efficiency, and real-time performance of robots within simulated human-centric environments (Page 2). The platform incorporates a diverse array of sensors into its modules to record, process, and plot data (Page 4). These sensors include: 2D lidar, 3D lidar, RGB, RGBD camera, IMU, sonar sensor, and foot contact sensor (Page 4). Furthermore, the platform extends its metrics suite with more social metrics (Page 4). The platform has been selected as the official hosting platform for competitions including the forthcoming SocialNav2025 workshop (Page 6).

User Study and Usability

The platform's performance was evaluated by generating the aforementioned benchmark and through a comprehensive user study, demonstrating significant improvements in usability and efficiency compared to previous versions (Page 1). A user study involving 20 participants from various universities assessed the platform’s strengths, weaknesses, overall impressions, suggestions for improvement, and ease of use (Page 7). Participants praised the new Isaac Gym integration and simplified wrappers, noting that they had previously avoided Isaac Sim due to limited documentation and complex setup (Page 7). Users also expressed favorable opinions about the web-based interface, stating that it is more intuitive and appealing than the previous JSON-based configuration workflow (Page 7).

Limitations

The primary limitation of Arena 5.0 is its increased computational requirements: photorealistic simulation incurs higher overhead and latency than lighter-weight simulators (Page 8). World generation is currently limited to three environment types (office, hospital, and warehouse), and expanding this coverage via globally sourced datasets is a priority for future work (Page 8). Human motion visualization remains dependent on pre-existing animation assets, and creating new animations requires specialized 3D modeling expertise (Page 8). Arena’s abstraction layer enables users to select the simulator best suited to their hardware and application needs (Page 8).

Arena 5.0 represents a substantive advance toward narrowing the sim-to-real gap (Page 8). This paper introduces Arena 5.0, a substantial enhancement over previous works designed to support the dynamic generation of 3D environments for training, testing, and benchmarking social navigation approaches (Page 8). The platform is completely open-source, providing transparency and extensive customization options through accessible APIs (Page 6). With its user-friendly graphical interfaces and faster processing times, Arena 5.0 has demonstrated significant improvements in functionalities and user experience (Page 8). Future developments aim to further expand the platform’s functionalities by integrating more advanced planning approaches, extend the world generation capabilities to include outdoor worlds and weather conditions, and multi-agent reinforcement learning approaches (Page 8). The forthcoming release of Arena-Web will replicate most of the platform’s features in a web-based application, enhancing its accessibility and ease of use for a global audience (Page 8).

--- Page 1 ---

Demonstrating Arena 5.0: A Photorealistic ROS2 Simulation Framework for Developing and Benchmarking Social Navigation Volodymyr Shcherbyna, Linh Kästner, Duc Anh Do, Hoang Tung, Huu Giang Nguyen, Maximilian Ho-Kyoung Schreff, Tim Seeger, Eva Wiese, Ahmed Martban, Huajian Zeng, An Tran Nguyen Quoc Hung Jona Kreutz Vu Thanh Lam Ton Manh Kien Harold Soh Equal Contribution 1National University of Singapore (NUS), Singapore 2Technical University Berlin (TUB), Germany 3Technical University Munich (TUM), Germany Abstract—Building upon the foundations laid by our previous work, this paper introduces Arena 5.0, the fifth iteration of our framework for robotics social navigation development and benchmarking. Arena 5.0 provides three main contributions: 1) The complete integration of NVIDIA Isaac Gym, enabling photorealistic simulations and more efficient training. It seamlessly incorporates Isaac Gym into the Arena platform, allowing the use of existing modules such as randomized environment generation, evaluation tools, ROS2 support, and the integration of planners, robot models, and APIs within Isaac Gym. 2) A comprehensive benchmark of state-of-the-art social navigation strategies, evaluated on a diverse set of generated and customized worlds and scenarios of varying difficulty levels. These benchmarks provide a detailed assessment of navigation planners using a wide range of social navigation metrics. 3) Extensive scenario generation and task planning modules for improved and customizable generation of social navigation scenarios, such as emergency and rescue situations. The platform’s performance was evaluated by generating the aforementioned benchmark and through a comprehensive user study, demonstrating significant improvements in usability and efficiency compared to previous versions. Arena 5.0 is open source and available at https://github.com/Arena-Rosnav I National University of Singapore (NUS), Singapore Technical University Berlin (TUB), Germany Technical University Munich (TUM), Germany

--- Page 2 ---

Fig. 2: Data flow of the Generation Stage and the Population Stage. The Generation Stage combines multiple SotA technologies to process text inputs into a floor plan image and room asset locations. 3DSGs are used as an intermediate data structure to divide the problem into a text transformation task solvable by an LLM, and a graph transformation task solvable by a spatial GNN. The Population Stage populates the floor plan’s asset zones with 3D models by employing the Asset Placer. A pre-built semantic vector Model Database is queried for a related model, which is arranged into the zone by a Fitter algorithm. After a final post-processing step, the end result is a finished environment consisting of 3D walls and models. To address these issues, we introduced world generation capabilities and social interaction modules in Arena 3, followed by a platform-wide migration to ROS2 and extended world-generation functionalities in Arena 4.0. Building on these foundations, Arena 5.0 integrates Isaac Gym to offer photorealistic simulations, thereby aligning with both research and industry requirements. By providing an enhanced world and scenario generation process, as well as a direct link to Isaac Gym, Arena 5.0 enables more realistic and comprehensive evaluations, bridging a critical gap in social navigation research and deployment. The main contributions of this work are as follows: • Full Integration of Isaac Gym into the Arena Framework: This includes a comprehensive ROS integration of Isaac Gym, facilitating seamless inter-operation with ROS-based systems. This integration aims to leverage Isaac Gym’s advanced simulation capabilities within the Arena framework to enhance the development and testing of autonomous robots in realistic environments. • Benchmarking: Establishment of rigorous benchmarks for evaluating the performance of social navigation algorithms. This includes metrics for assessing accuracy, efficiency, and real-time performance of robots within simulated human-centric environments. • Automated and Customizable Scenario Generation Pipelines: Development of flexible scenario generation pipelines that allow researchers to automatically create diverse and complex environments. This feature supports customization to address specific research needs and scenarios, thus accelerating the testing and refinement of navigation algorithms.

--- Page 3 ---

Fig. 3: System Design of Arena 5.0:

Improvements for AI systems

  1. Bold header: Full Integration of Isaac Gym into Arena Framework

This integration allows for photorealistic simulations and enables users to effortlessly generate social navigation scenarios that exhibit photorealistic behaviors, significantly lowering barriers to entry for researchers using advanced simulators like Isaac Sim.

  1. Bold header: Comprehensive Benchmarking Suite

The platform establishes a rigorous evaluation system by enabling the assessment of planners on a diverse set of generated and customized worlds and scenarios of varying difficulty levels, providing a detailed assessment of navigation planners using a wide range of social navigation metrics.

  1. Bold header: Automated and Customizable Scenario Generation Pipelines

The development includes automated and customizable scenario generation pipelines that support creating diverse environments, such as emergency situations, incorporating random behaviors—humans interacting unexpectedly with robots, to ensure algorithms are robust against unforeseen events.

  1. Bold header: Enhanced Sensor Modality Integration

By implementing wrapper classes for sensors like 2D lidar, 3D lidar, RGB, RGBD camera, IMU, sonar sensor, the system supports a wider variety of state of the art navigation approaches that are not limited to 2D sensors.

  1. Bold header: Generative AI-Driven World Creation

The system utilizes modules like 3dsg from text and MiDiffusion network to allow users to input a natural language prompt or upload a floorplan, which is then processed into a finished environment consisting of 3D walls and models, facilitating the intuitive creation of social scenarios.

  1. Bold header: Multi-Modal Social Metrics Evaluation

The platform extends metrics beyond traditional navigation performance by incorporating Social Metrics such as Time seen by pedestrians [s] and Time in personal space [s], allowing for a holistic evaluation of human-robot interaction quality.

Sources

Related papers