Package Delivery using Drones

High Precision location identification for multiple applications using deep neural networks by augmenting GPS with terrain knowledge

Principal Investigator (PI): Dr. Muhammad Bilal Kadri

Co-PI: Dr. Sameer Qazi

Karachi Institute of Economics and Technology (KIET)

Industrial Partner: Mr. Humayun Qureshi

Director Datalog

Funded by: National Center of GIS and Space Applications (NCGSA) 

Research Grant: 12.48 million PKR. 

Project duration: 1st May 2022 to 30th April 2024 

Project Brief

Accurate Positioning of a moving vehicle or UAV drone on a target location is essential in order to provide quality service to customers. Common problem with Uber/Careem, Bykea and others is that drivers are misguided because customer was not there. The root cause here is that GPS modules being used have inherent sensor error of 4-6 meters.

If a higher target location accuracy of 0.5-1 meter could be achieved then this whole market would open up to New Sensor Based Applications in Pakistan such as

1. Delivery of parcels via UAV’s  to homes or offices (huge cost & time saving)

2. Delivery of emergency food and medical supplies via UAV’s during disaster management (life saving)

3. Accurate location identification of  the injured so that rescue teams can be dispatched (life saving)

4. Accurate identification of locust infested areas for aerial spray (food saving / poverty alleviation)

5. Etc.

This project will develop algorithms for very accurate determination of any target location (down to +/- 1 meter). We will use deep neural networks augmented with Terrain Navigation Reference approach. As a Use Case we will design the algorithms and develop all the hardware; and then test it for a point-to-point parcel delivery service using UAV’s (quadcopter drones). Successfully developed product can be attractive for food delivery services, courier companies, large enterprises with multiple offices, large farms, university campuses, govt. agencies, etc. Our industrial partner has keen interest in technology products and can derive financial gains.  

Executive Summary

Retail industry has seen a boom in recent years. Thanks to online shopping stores and quick product delivery by various means. If the product delivery can be automated, the sales and consequently the revenue can increase exponentially. One of the major hurdles in automating the delivery mechanism is the inaccuracies associated with the GPS signals available in most handheld devices(Rahman, Nasihien et al. 2019). The performance of the localization system based on GPS degrades severely in urban environment which leads to false positive.

For accurate localization GPS signals should be augmented with some other sensor information to improve the localization accuracy(Shetty and Gao 2017, Song, Nuske et al. 2017). We propose to increase the localization accuracy (i.e. accurate longitude and latitude) provided by the GPS signal by augmenting it with the terrain information (e.g satellite images provided by google maps) using deep neural networks. With IoT enabled services the performance of the localization scheme as well as the delivery robot would be visible to the retailer as well as to the consumer.

The major objective of this research is to design and implement a product delivery system based on aerial robotics (quadcopters) by employing enhanced localization services based on machine learning. Specifically, the quadcopters will be utilized for product delivery services. The same idea can be applied in rapid response services for disaster mitigation, medicine delivery in harsh environment, booking a ride by taxi or other means, automobile tracking etc. 

A localization strategy will be adopted based on enhanced Terrain Reference Navigation (TRN)(Kim, Park et al. 2018) mechanism combined with the deep-learning strategy which aims to increase the location accuracy and hence the overall efficiency of the system. Additionally, the proposed aerial robotic based system has several advantages when compared with the conventional methods including (1) less man-power (2) it can provide faster transportation (3) lesser fuel consumption (4) consumer satisfaction and (5) raising the profitability.

The IMR Lab at PAF-KIET excels in design and development of ground and aerial robots. They have already designed an intelligent firefight robot. The project was funded by IGNITE (former ICTR&D fund). At IMR lab a customized quadcopter has already been designed that can navigate autonomously in an area. Localization scheme based on sensor fusion concept was proposed by researchers working in IMR Lab, for ground based robots. The scheme was published in a highly reputed Robotica journal published by University of Cambridge, UK. IMR Lab is currently working on a classified project related to “Advanced sensor fusion techniques for localization of UAV in harsh environment”.

Project Closure

The primary objective of this research project was to design and implement an advanced aerial product delivery system using quadcopters, enhanced with machine learning–based localization services. The system was envisioned to provide a robust and efficient solution for delivering critical supplies, particularly in remote or hard-to-reach areas. While the core focus was on medical supply delivery, the concept is adaptable to various rapid-response scenarios, including disaster relief operations, emergency medicine transport, ride-hailing services, and vehicle tracking.

The scope of work encompassed the development of an autonomous drone platform capable of safe, precise, and reliable navigation from dispatch to delivery. This included the integration of intelligent route planning, AI-assisted localization, and cloud-based mission management, ensuring the solution could operate with minimal human intervention and high operational efficiency.

The project successfully achieved all targeted milestones. An autonomous drone delivery system was developed and fully integrated with a cloud-based platform, enabling hospitals and healthcare facilities to rapidly deploy drones in response to medicine requests from remote dispensaries. Upon receiving a request, the system automatically generates GPS waypoints and dispatches the drone to the destination. For precision landing, the drone utilizes ground-based visual markers, ensuring safe delivery without the need for manual piloting. This approach enhances accessibility in areas with limited technical expertise, reduces response times, and improves delivery reliability.

The key deliverables include:

1. A fully functional autonomous drone delivery system.

2. Cloud-based mission management software for real-time operation.

3. AI-assisted localization and marker-based precision landing capability.

4. Demonstrated end-to-end delivery workflow for medicine distribution.