vEDGAR -- Can CARLA Do HiL?

TL;DR

This paper investigates using CARLA for Hardware-in-the-Loop testing of automated driving by introducing the vEDGAR framework, a GPU-accelerated, ROS2-based extension that enables EDGAR hardware to run automated driving software in a real-time-approximate HiL. It defines requirements, develops a service-oriented architecture, and implements a GPU-enhanced LiDAR, a configurable sensor stack, and a realistic actuation interface to enable testing end-to-end sense-plan-act pipelines. The authors validate the approach through a PoC evaluation across ROS2 interfaces, sensor real-time performance, and HiL scenarios, concluding that CARLA can support HiL testing but operates at its performance limits, with clear avenues for improvement. The work provides open-source tools and a pathway to integrating CARLA into the broader EDGAR-based research workflow, potentially stabilizing LoD-HiL workflows for automated driving research.

Abstract

Simulation offers advantages throughout the development process of automated driving functions, both in research and product development. Common open-source simulators like CARLA are extensively used in training, evaluation, and software-in-the-loop testing of new automated driving algorithms. However, the CARLA simulator lacks an evaluation where research and automated driving vehicles are simulated with their entire sensor and actuation stack in real time. The goal of this work is therefore to create a simulation framework for testing the automation software on its dedicated hardware and identifying its limits. Achieving this goal would greatly benefit the open-source development workflow of automated driving functions, designating CARLA as a consistent evaluation tool along the entire development process. To achieve this goal, in a first step, requirements are derived, and a simulation architecture is specified and implemented. Based on the formulated requirements, the proposed vEDGAR software is evaluated, resulting in a final conclusion on the applicability of CARLA for HiL testing of automated vehicles. The tool is available open source: Modified CARLA fork: https://github.com/TUMFTM/carla, vEDGAR Framework: https://github.com/TUMFTM/vEDGAR

Figures (9)

• Figure 1: The HiL setup investigated in this work consists of the x86 compute hardware and network setup from the EDGAR research vehicle, together with the vEDGAR tool that wraps CARLA for the HiL application
• Figure 2: comparison of real LiDAR from EDGAR (left) and the simulated LiDAR from vEDGAR at the inner city of munich for the Wiesn-shuttle demo.
• Figure 3: Schematic representation of the interfaces of the research vehicle EDGAR, [] indicates the ROS2 message class
• Figure 4: vEDGAR architecture applied for the HiL. Displayed are the various services that utilize a vEDGAR Plugin instance, the vEDGAR Server, and the CARLA Server, as well as how these services are distributed across the hardware. Blue arrows describe a data transmission via the network.
• Figure 5: The final LiDAR simulation achieved runs with 10Hz in a soft real-time condition using the GPU acceleration. It consists of four individual LiDAR sensors, two mechanical Ouster OS1 with 360 degrees coverage and two solid state Innovusion Falcon with a FoV of 120 degrees. The concatenated point cloud of the four LiDARs is visible in the image. Each sensor is simulated individually and merged in the automation softwarestack.