An Autonomous Obstacle Avoidance Vehicle System with TinyML for Last Mile Delivery

This project presents the design and implementation of a cost-efficient autonomous obstacle-evading vehicle, which is designed to operate in last-mile delivery with Tiny Machine Learning (TinyML) and Internet of Things (IoT) applications. Resolving the inefficiencies, high costs, and environmental impacts of traditional last-mile logistics, the study introduces an embedded system capable of making real-time obstacle detections and autonomous navigation in urban and semi-urban environments. The platform is based on an ESP32 microcontroller, incorporating ultrasonic sensing and Wi-Fi connectivity to enable data exchange with a remote server. An on-board lightweight FOMO (Faster Objects, More Objects) TinyML model is deployed on board to perform on-board object detection, and, therefore, allows decentralised, low-latency decision making without relying on cloud computation. Navigation and path planning modules were also put through extreme tests in varied environments to test the stability, responsiveness and delivery success rates. The experimental results demonstrate that the TinyML-based model is effective in obstacle detection, requiring minimal computational and energy costs, which justifies its suitability for use in resource-constrained settings. The suggested system, in turn, contributes to the growing body of literature on sustainable, edge-AI-based logistics by providing a scalable and energy-saving prototype of autonomous last-mile delivery in both smart and developing urban settings.

AI Cross-RefView PDF

Cite as: desci.ng.1308.2025

Uploaded on Mar 6, 2026, 11:35:25 AM

Autonomous VehiclesLast-mile deliveryFOMOESP32

Notes

The combination of FOMO object detection, ultrasonic sensing, and Wi-Fi IoT connectivity in a single low-cost embedded platform is a meaningful systems integration challenge. Demonstrating that it works reliably in varied environments gives other researchers a concrete foundation to build on.

Comments