Miniature Testbed for Validating Multi-Agent Cooperative Autonomous Driving
Hyunchul Bae, Eunjae Lee, Jehyeop Han, Minhee Kang, Jaehyeon Kim, Junggeun Seo, Minkyun Noh, Heejin Ahn · 2025
Read Miniature Testbed for Validating Multi-Agent Cooperative Autonomous Driving by Hyunchul Bae, Eunjae Lee, Jehyeop Han, Minhee Kang, Jaehyeon Kim, Junggeun Seo, Minkyun Noh, Heejin Ahn (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-11022.txt.
Introduction
Hyunchul Bae, Eunjae Lee, and their colleagues built CIVAT, a small-scale testbed for cars that drive and cooperate at one-fifteenth of real size. Cars that share data with each other and with smart road gear can drive more safely together, but no prior testbed gave the road gear full sensing and computing roles. CIVAT was built to fill that gap.
The setup pairs a scaled city map and sensor-equipped model cars with smart road units. Cars and road gear swap data over Wi-Fi through a shared publish-subscribe system. Tests on road-based sensing and crossing control show the testbed can support joint driving trials at low cost and low risk.
Catalog record
| CREATOR | Hyunchul Bae, Eunjae Lee, Jehyeop Han, Minhee Kang, Jaehyeon Kim, Junggeun Seo, Minkyun Noh, Heejin Ahn |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-11022
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Miniature Testbed for Validating Multi-Agent Cooperative Autonomous Driving Hyunchul Bae, Eunjae Lee, Jehyeop Han, Minhee Kang, Jaehyeon Kim, Junggeun Seo, Minkyun Noh, Heejin Ahn Abstract Cooperative autonomous driving, which extends vehicle autonomy by enabling real-time collaboration between vehicles and smart roadside infrastructure, remains a challenging yet essential problem. However, none of the existing testbeds employ smart infrastructure equipped with sensing, edge computing, and communication capabilities. To address this gap, we design and implement a 1:15-scale miniature testbed, CIVAT, for validating cooperative autonomous driving, consisting of a scaled urban map, autonomous vehicles with onboard sensors, and smart infrastructure. The proposed testbed integrates V2V and V2I communication with the publish-subscribe pattern through a shared Wi-Fi and ROS2 framework, enabling information exchange between vehicles and infrastructure to realize cooperative driving functionality. As a case study, we validate the system through infrastructure-based perception and intersection management experiments. I INTRODUCTION Autonomous driving technologies have advanced significantly over the past decades, and their widespread deployment is now imminent. However, most progress has focused on stand-alone vehicles, whereas cooperative driving—where vehicles and roadside infrastructure share information and coordinate their actions—remains a challenging but essential problem. The deployment of cooperative driving requires rigorous validation in realistic settings, but such validation continues to face substantial barriers in practice. To enable such validation, several full-scale testbeds have been established, including M-City [ 24 ] , K-City [ 11 ] , and AstaZero [ 16 ] . These facilities provide urban-like environments and diverse traffic scenarios, and are equipped with Vehicle-to-Everything (V2X) infrastructure for connected and automated vehicle testing. While they improve the prospects of commercialization, they also demand extensive space and massive investment, creating a significant entry barrier. For cooperative driving, realistic validation further requires smart infrastructure together with a sufficient fleet of autonomous vehicles to capture multi-agent interactions, which are often prohibitively costly and logistically demanding. As a result, researchers often turn to simulation environments, such as CARLA [ 5 ] , which provide safer and more cost-effective setups. However, simulations inherently struggle to replicate the physical realities of vehicle dynamics, sensor characteristics, and communication issues. To bridge the gap between full-scale testbeds and simulations, miniature testbeds have emerged as promising alternatives. These platforms enable controlled experiments at reduced cost and with lower entry barriers. Tab. I summarizes representative miniature testbeds, comparing them in terms of vehicle sensing, smart infrastructure (infrastructure with sensing, computation, and communication capabilities), reliance on motion capture systems, and supported interaction types (Vehicle-to-Vehicle (V2V) or Vehicle-to-Infrastructure (V2I)). Most existing platforms rely on onboard vehicle sensors and motion capture systems, while a few explicitly incorporate infrastructure-based sensing or V2I communication. ICAT [ 22 ] and MCCT [ 4 ] introduce infrastructure nodes with V2I communication, but still depend on motion capture systems and infrastructure without sensing capabilities. TABLE I : Scaled Vehicle Testing Platforms Testbed Vehicle Sensor Smart Infra Motion Capture V2V Comm. 1 V2I Comm. 2 Duckietown [ 15 ] ✓ - - - - UDSSC [ 21 ] ✓ - - - - Cambridge [ 8 ] - - ✓ - ✓ Go-CHART [ 9 ] ✓ - - - ✓ CPM Lab [ 10 ] ✓ - ✓ - ✓ ICAT [ 22 ] ✓ - - ✓ ✓ MCCT [ 4 ] ✓ - ✓ - ✓ CIVAT ✓ ✓ ✓ ✓ ✓ 1 Direct inter-vehicle communication. 2 Communication between vehicles and external agents. In this paper, we propose a novel 1:15-scale miniature testbed, CIVAT (Cooperative Intelligent V2X Autonomous Testbed) , designed for the validation of cooperative autonomous driving technologies. The platform integrates miniature vehicles equipped with onboard sensors and smart infrastructure supported by 3D LiDAR and a powerful edge computing module, thereby enabling advanced perception and decision-making. Recognizing that traffic environments will, for the foreseeable future, consist of a mix of connected autonomous vehicles (CAVs) and non-connected vehicles, our testbed supports both fully CAV and such mixed-traffic scenarios, providing a versatile platform for validating cooperative driving algorithms. Fig. 1 : Main Components of CIVAT The contributions of this study are as follows: • We design a 1:15-scale testbed for validating cooperative autonomous driving technologies, incorporating both vehicles and smart infrastructure. • We develop miniature vehicles equipped with onboard sensor suites and smart infrastructure equipped with 3D LiDAR and computation devices for perception, computation, and communication. • We present an intersection management case study using smart infrastructure’s detection with 3D LiDAR and V2I communication to coordinate autonomous vehicles under mixed-traffic conditions. The paper is structured as follows. Section II presents the design of the proposed testbed covering its main components: the map, vehicles and smart infrastructure. Section III details the software architecture. Section IV describes the case study of intersection management. Section V presents experimental validations on the testbed focusing on infrastructure-based detection and intersection management. Finally, Section VI presents the conclusions and future work. II TESTBED DESIGN Our testbed, CIVAT, is implemented by 1:15 scaling down a real-world four-way intersection. As shown in Fig. 1 , the main components of CIVAT include the map, vehicles, and smart infrastructure. We detail the design of each component and communication in this section. II-A Map Design Fig. 2 : Road Types and Segments We design the map shown in Fig. 2 to capture the complexity of urban traffic and the dynamic interactions between vehicles. We construct it at a 1:15 scale, which yields a physical size of 6 m \times 5.5 m corresponding to a real-world scale of 90 m \times 82.5 m. At the center, we implement an unsignalized four-way intersection with 2–3 lanes per approach (yellow region in Fig. 2 ) and introduce channelizing islands (red regions in Fig. 2 ) on the approach and departure lanes to induce realistic merging and diverging behaviors. To support continuous vehicle operation, we add a 2-lane beltway (blue regions in Fig. 2 ) that encircles the intersection and serves as a highway for high-speed driving. We model the road network as a graph, where each edge (black lines in Fig. 2 ) corresponds to a path segment. We store the waypoints along every segment in the memory of all agents to enable waypoint following. To replicate the frictional characteristics of asphalt, we print the map surface with material rated at 60 BPN (British Pendulum Number). II-B Vehicle Design II-B 1 Hardware Components Fig. 3 : Vehicle Components TABLE II : Hardware Components of Vehicle Component Item Price (US$) 1. Chassis & Servo motor ZD RACING S16 4WD 117 2. Single-board Computer Jetson ORIN NX 16GB 899 3. Custom PCB Regulator, Motor Driver 20 4. Microcontroller Unit NUCLEO-G431KB 13 5. Depth Camera RealSense D435 334 6. Encoder DC Motor M25N-2R-14 12V 18 7. IMU myAHRS+ 76 8. LiPo Battery Vega 22.2V 1300mAh 30 • The number in front of each component corresponds to Fig. 3 . Fig. 3 represents the overall hardware composition of the vehicle. The vehicle is based on a 1:15-scale chassis 1 1 1 Available at https://www.alibaba.com/product-detail/ZD-Racing-1-16-2-4G_1600493296884.html with Ackermann steering. It has a wheelbase of 17.5cm and a wheel track of 11cm, and its overall dimensions are 15\,\text{cm}\times 30\,\text{cm}\times 13\,\text{cm} . The onboard system includes the Jetson Orin NX for computation with GPU capabilities, equipped with an external Wi-Fi antenna for V2X communication.