Project Abstract

Developing robust Advanced Driver-Assistance Systems (ADAS) using Reinforcement Learning (RL) benefits greatly from high-fidelity simulators like BeamNG.tech. However, utilising its capabilities effectively through standard scripting presents significant workflow hurdles, including cumbersome setup, difficult real-time monitoring, and rigid configuration, ultimately hindering research productivity. This research proposes an integrated graphical desktop environment specifically designed to address these workflow challenges by offering a visual interface tightly coupled with the simulator. The goal is to streamline the process of developing, training, and monitoring sensor-driven RL agents within BeamNG.tech — without relying solely on code. The methodology involved designing and implementing the desktop application using a client-server architecture (React/Electron frontend, Python/FastAPI backend communicating via REST/WebSockets) to manage BeamNG.tech interaction via beamngpy. The tool’s capabilities were validated through an RL driving case study using the simulator’s RoadsSensor (simulated sensor which gives geometric and semantic data of the road). Key findings demonstrate that the GUI-based workbench enhances the training workflow of the driving agent compared to traditional script-based approaches. It provides an intuitive interface for configuration, integrated real-time visualisation of RL metrics and sensor data, and enables faster iteration cycles — all within a unified, accessible application. The result is a fully functional BeamNG.tech “workbench” software tool that brings powerful simulation features into an easy-to-use graphical environment. While this study focused on a reinforcement learning use case, the flexible, GUI-driven nature of the application makes it equally suitable for broader ADAS experimentation and intelligent vehicle research beyond RL. Ultimately, this work contributes a novel, validated graphical software tool and workflow solution that lowers the barrier to entry and enhances efficiency for researchers conducting sensor-driven ADAS, RL, or general simulation-based experiments in the powerful BeamNG.tech environment — making complex research tasks more visual, interactive, and accessible.

Keywords: Autonomous Driving, ADAS Development GUI, Artificial Intelligence Machine Learning

 Conference Details

Session: A

Location: Sir Stanley Clarke Auditorium at 11:00 13:00

Markers: Matt Roach, Xianghua Xie

Course: BSc Computer Science 4yr FT

Future Plans: I’m looking for work