Explore the technology behind autonomous driving

In just a few years

Autonomous driving has gone from a futuristic dream in high-level meetings to a

Developing into a current reality that everyone is talking about

 

As technology matures

Personal and public transportation will undergo a sea change

Finally, driverless cars

Or human drivers will be excluded

No more on the road

Drowsy, low-energy and distracted drivers

However, with the emergence of autonomous driving

In addition to excitement and anticipation, there are many questions:

“Is autonomous driving safe?”

"We can really let go of the steering wheel.

Don’t you need to pay attention to the road conditions yourself? "

What is the technology behind autonomous driving? Why will this technology eventually be trusted? This article will analyze how we can embrace fully autonomous driving from the perspective of artificial intelligence and memory. Feel free to share your views on autonomous driving at the end of the article to win a small gift of "black technology".

For a car to drive itself, it must constantly understand the environment around it—first by sensing (recognizing and classifying information) and then acting on that information through the car’s automated/computerized control systems. Self-driving cars require safe, reliable, and highly responsive solutions that can make split-second decisions with detailed knowledge of the driving environment. Understanding the driving environment requires massive amounts of data captured by the car’s numerous sensors, which are then processed by the car’s autonomous driving computer systems.

In order for a car to truly be able to drive without a human in control, the AI ​​network must first be trained extensively to understand how to see around it, how to interpret what it sees, and make the right decisions in any imaginable traffic situation. The computing performance of self-driving cars is comparable to some of the highest-performance platforms that were available only a few years ago.

Autonomous vehicles are expected to contain more lines of code than any other software platform to date. By 2020, a typical car is expected to contain more than 300 million lines of code and will contain more than 1 TB of storage, requiring more than 1 TB per second of memory bandwidth to support the computational performance required by autonomous driving platforms.

The artificial intelligence systems of self-driving cars require a continuous, uninterrupted flow of data and instructions in order to make real-time decisions based on complex data sets. Today, there are successful self-driving vehicles on the road, but the success of these early cars is the result of driving the same routes repeatedly for many days in a row, learning the details of each route and generating high-resolution maps as a key part of the self-driving navigation system.

With less need to identify routes, the car's computer can focus on potential real-time hazards such as traffic, pedestrians, etc. This generally limited range of operation is called "geofencing" and reflects the approach taken by early self-driving cars. While geofencing can provide a solution that works well on restricted routes, a self-driving car that relies heavily on geofencing in one part of the world may not work properly in another part of the world.

Whether it is the memory subsystem related to sensor fusion processing, path planning, or the storage subsystem related to the black box data recorder, from solid-state drives (SSDs) to NAND flash, from NOR flash to low-power DRAM and GDDR6, various memory and storage devices play a vital role in bringing us closer to the dream of the future - we can answer emails, answer Skype calls or watch our favorite shows in the car. At the same time, our self-driving car can safely take us to our destination through the best route.

Human drivers can drive tired or distracted, respond slowly, be indecisive, have poor judgement and make human errors; self-driving cars have a 360-degree field of view, a maximum observation range of 200 yards, 24/7 full-time focus, are never distracted, and can respond faster and more accurately. Vehicle-to-vehicle communications also allow other cars to drive safely.

Imagine a car slams on the brakes on a busy highway. By introducing communications between cars and other cars and infrastructure (collectively known as V2X), this single event of braking can be wirelessly transmitted to all vehicles in front and behind, allowing them to understand the current situation in a timely manner and actively slow down and brake to avoid accidents.

90% of fatal traffic accidents in the United States in 2017 were caused by human error. Humans are easily distracted and can make quick decisions when faced with unexpected dangers. Computers, on the other hand, are not distracted and react more consistently and quickly than human drivers.

Safety of self-driving cars is a top concern, and understandably so. The focus on safety goes far beyond the redundancy built into the hardware system to avoid erroneous decisions, and includes an associated infrastructure that enables vehicles to communicate with each other and with their surroundings. This wirelessly interconnected computing subsystem with hardware redundancy is regulated by legislatures whose purpose is to mandate a level of safety that is directly related to the level of automation.

As a regulatory measure for the development and deployment of autonomous driving technology, NHSTA has developed a series of levels to determine the degree of control that humans and computers have over vehicles. The range is divided into: Level 0 (automation), Level 1 (driver assistance), Level 2 (partial automation, the driver needs to keep one hand on the steering wheel), Level 3 (conditional automation, the driver may need to be taken over at any time), Level 4 (high automation) and the highest Level 5 (full automation). Currently, most ADAS solutions are level 2 and are based on computer hardware using relatively mature and low-bandwidth memory devices.

As driverless cars reach higher and higher levels of automation, the importance of memory technology from a safety and performance perspective has moved from the back seat of the car to the front seat. Historically, personal computers have been considered the driver of memory technology, but now the automotive industry will be the main driver of memory technology in the future. This has been demonstrated by some of the leading autonomous platforms today.

NVIDIA's recently announced state-of-the-art Pegasus computing platform is specifically developed for autonomous driving and is based on the industry's highest-performance cutting-edge DRAM technology. Overall, the Pegasus platform provides more than 1 TB per second of memory bandwidth to support Level 5 autonomous driving performance.

Micron has been a recognized leader in automotive memory solutions, graphics memory GDDR5 and GDDR6. The bandwidth associated with GDDR6 memory enables higher levels of autonomy in the actual memory footprint that can be deployed in the car. An autonomous computing platform with abundant memory bandwidth will be able to support the continued development and improvement of autonomous driving algorithms.

The continued development of autonomous vehicles means that there will be multiple iterations of capabilities over the next decade. This will require careful management of human-machine interactions, ensuring that drivers clearly understand what level of autonomy is available at a given time and what their responsibilities are for "hands-on driving" and "close attention" operations.

GDDR6 is a foundational technology that provides the essential memory bandwidth to drive the AI ​​compute engines that enable self-driving cars to act responsibly and ensure safety in compliance with industry safety standards set by NHSTA. GDDR6 is the highest performance memory technology available today and is capable of operating in the high temperatures and harsh conditions associated with automotive.

Artificial intelligence is the key technology to realize autonomous driving

The extreme computing performance required for AI-based self-driving cars

Innovative memory and storage systems are needed to handle and store large amounts of data

This data is necessary for computers to make human-like decisions

As self-driving cars demand faster memory

Micron has been deeply involved in the automotive industry for more than 25 years

It will enable it to maintain its leading edge

Providing the right level of performance