III: Medium: Collaborative Research: Integrating Large-Scale Machine Learning and Edge Computing for Collaborative Autonomous Vehicles

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

Monitoring of possible hazards and disasters are crucial for mitigating their effects on the physical environment or to humans. The unmanned Aerial Vehicles (UAVs) have been successfully used in surveillance systems, also for many other applications such as monitoring infrastructure, vegetation growth, coastline, traffic, etc. Due to the widespread applications, a higher level of intelligence and autonomy is required to ensure safety and operational efficiency. The emerging high-resolution sensors and deep learning techniques hold great promise for autonomous UAVs. However, the unprecedented scale and complexity of sensing data (such as aerial images) have presented critical computational bottlenecks requiring new concepts and enabling tools. To address these challenges, this project focuses on designing principled large-scale machine learning, edge computing systems, energy efficient algorithms and tools that are used to achieve the real-time prediction, utilize cloud and edge computing resources, advance data-driven model-based approaches, assure the safe and agile collaborative vehicles navigation. These results address the challenges in decision support and data revolution and lead to the next generation collaborative autonomous systems. The research objective of this project is to address the computational challenges in the innovative real-time and intelligent collaborative autonomous vehicles. A novel large-scale machine learning and edge computing framework is developed to integrate the emerging key computational techniques, including fast deep learning optimizations, asynchronous federated learning, cross domain deep learning model compression, hierarchical edge computing, and collaborative autonomous aerial and ground vehicles. Unlike most existing systems that perform big data analysis in central servers or clustering for offline learning, this project provides promising new directions to the real-time analysis of high-throughput sensor data by addressing the critical embedded device data analysis issues including efficiency, scalability, distributed computing, energy saving, and space reduction. The research project combines rigorous theoretical analysis and emerging application studies, and contributes to both academic research and potential commercialized products. Such unique capabilities enable new computational applications in a large number of research areas. It advances and thus extends the relationship between engineering innovation and computational analysis. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.