From delivery truck to research platform
The project began when two electric vehicles were delivered to NED University as the foundation for a research-driven autonomous driving platform. The goal was to transform them from conventional EVs into robotic testbeds capable of sensing the environment, localizing themselves, and executing controlled autonomous motion.
That meant the work was never just "add sensors." It meant understanding the vehicle's own specification well enough to modify it safely, standing up charging and testing infrastructure from scratch, integrating state-of-the-art localization and perception hardware, and validating all of it through live motion tests. The effort combined hardware integration, software pipeline setup, and iterative system testing to move from concept toward real-world autonomous driving behavior — and it ended in a successful autonomous drive test, after which the platform continued to evolve under a new lead once the original researcher moved on.
02An EV is not a robot yet
The core challenge was turning a regular electric vehicle into an autonomous platform with reliable perception and navigation. That required careful integration of GPS/RTK localization, inertial and visual tracking, 3D mapping sensors, steering actuation work, and robotics middleware to coordinate all of it — while managing real-world constraints like mounting stability, calibration quality, signal reliability, and safe operation in a dynamic environment.
On top of the technical load, the project carried a real coordination problem. A team of interns was brought in to develop and refine the steering mechanism while perception, mapping, and validation work ran in parallel. The hard part wasn't inventing a system — it was making it robust enough to run in a real vehicle, under real sensing and safety constraints, with several workstreams converging on the same chassis at once.
03Procurement to autonomous motion
The project started with procurement and setup: the vehicles were unpacked, inspected, and studied for their mechanical and electrical characteristics before any sensor touched them. Once that baseline was understood, work moved into vehicle preparation, charging-station setup, and hardware calibration, before expanding into a multi-layer autonomous architecture combining precise localization, environmental awareness, and actuation support.
Steering absorbed the most sustained effort. Interns and collaborators worked through iterations of a mechanism that could translate sensor feedback into physical steering motion, while the broader system integrated real-time positioning and situational awareness in parallel. The platform was assembled around a ROS-based architecture used for sensor fusion, robot-state monitoring, and live data visualization in tools like RViz — the connective tissue that let independently-developed subsystems actually talk to each other.
This was applied research, not an academic exercise: the goal was a real vehicle driving autonomously in a practical lab environment, and a usable platform other researchers could keep building on afterward.
04Four systems, one vehicle
Localization
A Reach RS2+ multi-band RTK GNSS unit anchored the vehicle's position, with ROS nodes configured to stream live GNSS data and verify positional accuracy through topic-level validation before it was ever trusted for motion control.
Perception
Intel T265 tracking cameras and ZED 2i depth cameras supplied visual and inertial perception, giving the vehicle a sense of its own motion and the space immediately around it, independent of GNSS.
3D sensing
A RoboSense LiDAR unit was roof-mounted and integrated into ROS for real-time 3D environmental mapping — a placement that demanded careful calibration to account for vibration and mounting-induced offset.
Autonomy stack
ROS sat at the center of it all, unifying sensor streams and robot state so navigation logic and live visualization could run against one consistent picture of the vehicle and its environment.
Results and lessons
The project demonstrated that autonomous driving isn't primarily a software problem — it's a full systems-integration challenge spanning hardware procurement, mechanical mounting, calibration, sensor synchronization, and runtime validation. The successful autonomous drive test confirmed the platform could operate in a practical driving scenario, not only in simulation.
The broader lesson was that autonomy depends on disciplined engineering: correct sensor placement, accurate calibration, robust mounting, consistent ROS integration, and iterative testing. Those are exactly the details that make research vehicles hard to build and hard to trust — and exactly the areas that define real progress in autonomous systems. The project left a strong foundation for continued research, even after the work paused when the original lead researcher moved on from the role.