Implementation of deep learning control for a quadcopter

Project Title

Implementation of deep learning control for a quadcopter

Project Area of Specialization Artificial IntelligenceProject Summary

Nowadays, Unmanned Aerial Vehicles are considered to be one of the fastest-growing technologies in the world. UAVs such as quadcopters are emerging in all the technical as well as non-technical fields. Hence, it makes them an influential and robust tool which is contributing a lot to serving humankind. Moreover, they are also assisting in the evolution of an improved lifestyle.
Momentous and substantial developments have been done in the process of vehicular autonomy over the past recent years. It all became possible with the escalation and augmentation of deep learning techniques and approaches in computer-based applications. UAVs are now rapidly approaching the eventual objective of near-complete autonomy. Quadcopters based on PID controllers have several limitations discovered over the years. For this purpose, other advancements in technology such as deep learning or reinforcement learning are being implemented in the quadcopters.
This report proposes a quadcopter whose flight controller is designed on the basis of deep neural networks. A real-time quadcopter flight experiment will be performed to check the speed, precision, accuracy, and other such attributes. Moreover, this thesis is based primarily on the deep learning control which is modeled and simulated on GymFC and Neuroflight. The use of neural networks has presented extremely impressive and remarkable results over the recent years. So, we have demonstrated, through this report, that our flight controller is better in terms of speed, performance, and precision.

Project Objectives

The objectives and goals of our project are to:
? Develop an understanding of the Quadcopter system
? Study of the deep-learning techniques to train neural networks
? Bring improved latency and throughput using MCU STM32F7x2 based F722-STD Flight controller.
? Develop a toolchain for optimizing, compiling, and integrating trained neural networks on MCU.

Project Implementation Method

.

Benefits of the Project

Deep learning, when applied to data science, can offer better and more effective processing models. Its ability to learn unsupervised drives continuous improvement in accuracy and outcomes. It also offers data scientists more reliable and concise analysis results

Technical Details of Final Deliverable

.

Final Deliverable of the Project HW/SW integrated systemCore Industry AgricultureOther Industries Food , Manufacturing , Transportation , Media , Security Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Industry, Innovation and Infrastructure, Partnerships to achieve the GoalRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	32450
Matek F722 Flight controller	Equipment	1	10000	10000
Brushless motors	Equipment	4	800	3200
ESC	Equipment	1	8000	8000
Power Distribution Board	Equipment	1	500	500
Radio Transmitter	Equipment	1	1000	1000
LiPO battery	Equipment	1	2000	2000
Barometer	Equipment	1	500	500
GPS module	Equipment	1	1000	1000
IMU Sensor	Equipment	1	2000	2000
IR sensors	Equipment	5	50	250
Propellors	Equipment	4	1000	4000