| KEYWORD |
Area Engineering
Slimmable Neural Networks for Perception aboard Nano-Drones
Thesis abroad
keywords ARTIFICIAL INTELLIGENCE, CONVOLUTIONAL NEURAL NETWORKS, DEEP LEARNING, DEEP NEURAL NETWORKS, DRONES, EMBEDDED SYSTEMS, ENERGY EFFICIENCY, FIRMWARE DEVELOPMENT, LOW POWER, MICROCONTROLLERS, ROBOTICS, SOFTWARE, UAV
Reference persons DANIELE JAHIER PAGLIARI
External reference persons Dr. Marco Levorato (UC Irvine), Dr. Alessio Burrello (Politecnico di Torino), Dr. Beatrice Alessandra Motetti (Politecnico di Torino)
Research Groups DAUIN - GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA, ELECTRONIC DESIGN AUTOMATION - EDA, GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA
Thesis type EMBEDDED SOFTWARE DEVELOPMENT, EXPERIMENTAL, RESEARCH, SOFTWARE DEVELOPMENT
Description If interested in this project, please write me an email attaching your CV with a transcript of your exam scores.
This project considers the task of perception by means of deep neural networks deployed aboard nano-drones. The constrained sensory and computational capacities provided within the power constraints on these tiny robotic platforms (drones with a diameter of approximately 10 cm) imply complex challenges for the real-time execution of deep learning models onboard. The goal of the thesis is the development of an adaptive perception network, that is able to adjust the amount of computation for the inference conditioned on the complexity of the input frames. In particular, the student will explore the application of approaches based on slimmable neural networks, that allow to adjust the width of the layers dynamically at runtime. After the design and the finalization of the perception network, the second milestone of the project will be the actual deployment of the network on a GAP8 System-on-Chip, the same SoC available on the Crazyflie 2.1 nano-drone to run the onboard intelligence.
See also screenshot 2024-07-17 at 17.00.04.png https://arxiv.org/abs/1812.08928
Required skills Proficiency in Python is required. Familiarity with Deep Learning and the PyTorch library is a plus. Furthermore, experience with C programming is desired.
Notes The Thesis is in collaboration with Prof. Marco Levorato in University of California, Irvine (USA).
Deadline 17/07/2025
PROPONI LA TUA CANDIDATURA