Veg Delivers GPS-Free Autonomous Drone Navigation with SLAM on a Raspberry Pi 4, Arduino Nano
Low-cost hardware delivers impressive on-drone capabilities, including fault handling, object detection, and facial recognition.
Researchers from Guru Gobind Singh Indraprastha University and the Delhi Skill and Entrepreneurship University (DSEU) claim to have developed a new approach for autonomous drone navigation without relying on Global Navigation Satellite Systems: वेग, or Veg.
"We present an autonomous aerial surveillance platform, वेग, designed as a fault-tolerant quadcopter system that integrates visual SLAM [Simultaneous Localization and Mapping] for GPS-independent navigation, advanced control architecture for dynamic stability, and embedded vision modules for real-time object and face recognition," the team explain of the project, also known as Veg or Speed if literally translated from the original Sanskrit to English. "The embedded vision system, based on a lightweight CNN [Convolutional Neural Network] and PCA [Principal Component Analysis], enables onboard object detection and face recognition with high precision."
When the team says "lightweight," they're not joking: the prototype system runs entirely on-drone, using only a low-cost Raspberry Pi 4 Model B single-board computer and an Arduino Nano-compatible microcontroller. The performance, admittedly, is limited, with the CNN delivering around two object detection frames per second with a 90 per cent precision rate — but by keeping the vision part of the system independent from the control loop, the team says the system can still respond in real time.
The team's platform is based on a cascaded control architecture, with a low-level linear quadratic regulator handling stabilization of pitch and roll with a proportional-derivative (PD) controller running in an outside loop for trajectory tracking. The open-source ORB-SLAM3 library is used for pose estimation with six degrees of freedom, closing the loop and providing drift correction — while a Dijkstra-based algorithm provides autonomous navigation over even complex terrain, including indoor mazes.
"Built on low-cost hardware and open-source components," the researchers claim, "Veg demonstrates that GPS-independent aerial autonomy can be achieved without sacrificing mission-critical features such as emergency handling, object detection, and onboard face recognition. Comprehensive simulations validate Veg's ability to track trajectories, handle rotor loss, and conduct aerial surveillance with minimal latency. With no reliance on external computation or GPS, the system remains self-contained and field-deployable in complex environments such as indoor arenas, industrial facilities, or urban canyons."
A preprint of the team's paper is available on Cornell's arXiv server under open-access terms, while the firmware and software for the project has been published on GitHub under the permissive MIT license. The only close-up picture of the drone itself, however, is a poor generative AI rendering — so anyone looking to try out Veg for themselves is advised to double-check all the claims made in the paper, as it has not been peer reviewed.
