Artificial intelligence driving autonomous vehicle development

Artificial intelligence driving autonomous vehicle development

Autonomous driving is one of the key application areas of artificial intelligence (AI). Autonomous vehicles (AV) are equipped with multiple sensors, such as cameras, radars and lidar, which help them better understand the surroundings and in path planning. These sensors generate a massive amount of data. To make sense of the data produced by these sensors, AVs need supercomputer-like, nearly instant processing capabilities. Companies developing AV systems rely heavily on AI, in the form of machine learning and deep learning, to process the vast amount of data efficiently and to train and validate their autonomous driving systems.

AI, machine learning, deep learning

Although AI, machine learning, and deep learning are sometimes used interchangeably, they don't refer to the same concepts. In the simplest terms, AI is a branch of computer science that covers anything related to making machines smart. So, when a machine completes tasks based on a set of rules that solve a problem, such intelligent behavior can be described as AI. Machine learning and deep learning are ways to create or train AI. Machine learning is the study of structured data and algorithms that the machine uses to perform a specific task without specific instructions. Machine learning is an application of AI that enables systems to learn and improve from experience.

Deep learning is a subset of machine learning, or the next evolution of machine learning. Deep learning is inspired by information processing patterns found in the human brain. It leverages complex neural networks that extract more detailed features as the neural network continues to learn and evaluate its input data. Deep learning can be supervised or unsupervised: supervised learning relies on labeled training data, whereas unsupervised learning utilizes less structured training sources.

Companies developing AV technology are mainly relying on machine learning or deep learning, or both. A major difference between machine learning and deep learning is that, while deep learning can automatically discover the feature to be used for classification in unsupervised exercises, machine learning requires these features to be labeled manually with more rigid rulesets. In contrast to machine learning, deep learning requires significant computing power and training data to deliver more accurate results.

In the past few years, deep learning has helped companies accelerate AV development programs. These companies are increasingly relying on deep neural networks (DNN) for more efficient processing of sensor data. Instead of manually writing a set of rules for an AV to follow, such as "stop if you see red", DNNs enable AVs to learn how to navigate the world on their own using sensor data. These algorithms are inspired by human brain, implying they learn by experience. According to a blog by NVIDIA, a specialist in deep learning, if a DNN is shown images of a stop sign in varying conditions, it can learn to identify stop signs on its own. However, companies developing AVs are required to write not just one but an entire set of DNNs, each dedicated to a specific task, for safe autonomous driving. There is no set limit of how many DNNs are required for autonomous driving; the list is actually growing as new capabilities are emerging. To actually drive the car, the signals generated by the individual DNN must be processed in the real time, which is done by high performing computing platforms.

Early use of AI in autonomous driving

The first use of AI for autonomous driving goes back to the second Defense Advanced Research Projects Agency (DARPA) Autonomous Vehicle Challenge in 2005, which was won by the Stanford University Racing Team's autonomous robotic car 'Stanley'. The winning team, led by Sebastian Thurn, an associate professor of computer science and director of Stanford Artificial Intelligence Laboratory, attributed the victory to use of machine learning. Stanley was equipped with multiple sensors and backed by custom-written software, including machine learning algorithms, which helped the vehicle find the path, detect obstacles and avoid them while staying on the course. Thurn later led the 'Self-Driving Car Project' at Google, which eventually became Waymo in 2016.

Read more:  https://ihsmarkit.com/research-analysis/artificial-intelligence-driving-autonomous-vehicle-development.html