How HPC, AI, and IOT Drive the Future of Smart Cars

2020-01-10

When many people think about the Internet of Things (IoT) and smart cars from a computing perspective, the first thing that comes to mind is the challenge of analyzing and feeding data into remote systems with small processors. This is certainly an important part of the IoT, but many complex applications require supercomputing resources for development, large-scale analysis, and advanced machine learning.


Connected cars require in-depth research in new areas including remote sensing and artificial intelligence. These efforts require significant computing and research resources, and not all car manufacturers have the capability to build the next generation of smart cars.


With this in mind, an ideal research center for connected and autonomous vehicles of the future would have computing capabilities scalable to supercomputer levels, while also supporting research work for developing smaller IoT or sensor devices. Such a center would also need to provide world-class emphasis on multidisciplinary research to support a wide range of projects in terms of funding, available technology, and collaboration.


Michigan University covers all these key elements, particularly in terms of research authorization. Additionally, Michigan University has world-class supercomputing resources to drive the future of connected car research. From exploring microdevices that connect, analyze, and transmit individual car information, to machine learning algorithms trained on its GPU-supported systems, to large-scale simulations for more efficient traffic flow, the university is always at the forefront of intelligent sensing vehicles.


We will explore the importance of high-performance computing (HPC) connectivity and why it is crucial for the success of this automotive and IoT research. However, first, it is worth noting some of the projects that have been enabled by the university's MCity initiative.



Smart Cars, Smart Cities


MCity is a wide-ranging interdisciplinary initiative at the University of Michigan aimed at developing smarter cities with a focus on the role of smart connected cars. These are key elements of smart cities as they are expected to reduce traffic congestion, improve safety, and even reduce carbon emissions by reducing idle time due to traffic congestion, among other benefits.


The initiative is driven by numerous research groups at the University of Michigan, involving fields such as economics, public health, engineering, and cybersecurity. Industry partners include Honda, Ford, General Motors, Toyota, as well as insurance companies including State Farm, and technology partners.


One of MCity's main features is a 32-acre testing facility with over 16 acres of road area, simulating scenarios that drivers encounter, including hazards caused by simulated tree cover and GPS black spots. Many real-world and simulated testing conditions for autonomous vehicles have been researched here.


As smart cars roam the testing infrastructure, countless research projects are collecting data. One team focuses on computer vision algorithms that extract clues almost in real-time from human observations, while another team focuses on pedestrian positioning systems to classify their behavior for more accurate predictions in cars. Others refine intelligent transportation systems using a large amount of travel data, while some use data to better understand how drivers interact with warning systems inside connected vehicles.


All these research projects use the computing resources provided by the university. At the core of these and future plans is the upcoming "Great Lakes" supercomputer, planned and tested in collaboration with Dell EMC.



Computing Infrastructure


Given the needs of the MCity initiative and the various research departments across the campus, Dell EMC and the University of Michigan had to collaborate closely to determine the factors cluster "Great Lakes" would provide the greatest value for the widest range of applications.


One of the growing needs for MCity, in particular, is the resources to leverage AI workloads to extend traditional modeling and simulation. Training a deep learning model to recognize road signs, pedestrians, and hazards, for example, can be a computationally intensive task, and the massive parallelism of GPUs is well-known to be effective in achieving this goal.


The "Great Lakes" supercomputer will feature 44 of Nvidia's highest-performing GPUs, the V100, with Nvidia's TensorCore components designed to accelerate deep learning training problems by leveraging dense multiply/accumulate operations for input and layer processing. These will be accessed for training data for AI models deployed in devices inside smart and connected vehicles to provide near real-time intelligence. Without high-performance, scalable GPUs installed in the "Great Lakes" machine, the efficiency of researching these algorithms and deploying general AI frameworks like TensorFlow and PyTorch for autonomous vehicle inference would be greatly diminished.


Dell EMC worked closely with networking partner Mellanox to provide another crucial feature for large-scale research work like MCity. Many applications on the test track (not to mention the university's 2,500 other active users) will require high bandwidth to sustain the performance of GPUs and CPU cores. In addition to being at the forefront of autonomous driving and smart city research, the University of Michigan has become a global leader in network performance, being the first production customer with Mellanox HDR 200 Gb/s Infiniband. This means projects that require multiple nodes, including those enabled with GPUs, will have access to the fastest data transfer speeds on Earth (thus obtaining application performance results).



Strong Partnerships


The success of the MCity initiative depends on strong collaborations between academia, industry, and technology providers.


The core technologies inside connected cars will drive future developments, and supporting the development of these applications is a top priority. The "Great Lakes" cluster is the result of collaboration that examined the efforts of MCity and the networking needs across the entire campus to build the perfect system for IoT, HPC, and AI research at the University of Michigan.


Palen added that Dell EMC has proven to be an outstanding partner as his team seeks to serve both traditional HPC workloads and emerging applications for connected and smart car research. For example, without the computing power to model and develop concepts for IoT devices and how they communicate with each other over networks, the scope of the MCity initiative would be greatly limited.


For most aspects crucial to autonomous vehicles, artificial intelligence is a primary feature; from computer vision applications to inferring human interactions with smart cars, the program would be limited without world-class supercomputers equipped with AI-optimized GPUs and the fastest network for research to date.


This is an exciting time to see how various domains in computing infrastructure are converging. From accelerated supercomputers like "Great Lakes" to microdevices that transmit signals and data to other smart car networks, this convergence will continue to create intelligent cars, smarter cities, and pave the way for sustainable intelligent machines.

RECENT AI NEWS

RECENT AI TOOLS