Problems of Self-Diving Vehicles and How to Solve Them – Thought Leaders

Autonomous vehicles require more than simple artificial intelligence. A self-driving car receives data from various sources such as sonars, cameras, radars, GPS, and lidars allowing it to navigate in any environment. Information from these devices should be processed quickly, and data volumes are massive.

The information from sensors is processed not only by the car’s computer in real-time. Some data is sent to peripheral data centers for further analysis. And then, through a complex hierarchy, it is redirected to various clouds.

The AI that the vehicle is endowed with is crucial, but also the processing capabilities of onboard computers, peripheral servers, and the cloud. The speed of sending and receiving data by the car, along with low latency, are both also very important.

Data Volume Problem

Even ordinary cars, with a driver behind the wheel, are generating more and more data. Self-driving cars can generate approximately 1TB of data per hour. This amount of data is simply gigantic. And it represents one of the barriers to the mass adoption of autonomous driving.

Unfortunately, all the data of a self-driving car cannot be processed in the cloud or peripheral data centers as this introduces too much delay. Even a 100-ms delay can make the difference between the life or death of a passenger or pedestrian. The car must respond to emerging circumstances as quickly as possible.

To reduce the delay between receiving information and responding to it, part of the information is analyzed by the onboard computer. For example, new Jeep models are equipped with an onboard computer with 25-50 processing cores that serves cruise control, blind-spot monitor, obstacle warning, automatic braking, etc. Vehicle nodes communicate with each other via an internal network. It also fits into the concept of peripheral computing if we consider the onboard computer as a peripheral node of the network. As a result, unmanned vehicles make up a complex hybrid network that combines centralized data centers, the cloud, and many peripheral nodes. The latter are located not only in cars but also in traffic lights, control posts, charging stations, etc.

Such servers and data centers outside the car provide all possible assistance with autonomous driving. They allow the car to “see” beyond the range of its sensors, coordinate the load on the road network, and help make optimal decisions.

Interaction With Each Other and Infrastructure

GPS and computer vision systems provide self-driving cars with information about their location and immediate surroundings. However, the range of the calculated environment is constantly increasing. Still, one car can only collect a limited amount of information. So, data exchange is absolutely necessary. As a result, each vehicle can better analyze driving conditions based on the more significant data set collected by the autonomous vehicle fleet. Vehicle-to-vehicle (V2V) communication systems rely on mesh networks created by vehicles in the same geographic area. V2V is used to exchange information and send signals to other vehicles, such as distance warnings.

V2V networks can be extended to share information with traffic infrastructure such as traffic lights. It is already appropriate to talk about V2I (vehicle-to-infrastructure) communication here. V2I standards continue to evolve. In the US, the Federal Highway Administration (FHWA) regularly issues various V2I guides and reports to help improve the technology. The benefits of V2I extend far beyond security. In addition to enhancing safety, vehicle-infrastructure technology provides advantages in terms of mobility and interaction with the environment.

Drivers who go the same route every day remember all the potholes on the road. Self-driving cars are also constantly learning. Self-driving cars will upload available helpful information to peripheral data centers, for example, integrated into charging stations. Charging stations will rely on artificial intelligence algorithms that will help analyze the data received from cars and offer possible solutions. Through the cloud, this data will be transmitted to other unmanned vehicles in the common network.

If this model of data exchange between all self-driving cars really materializes in a few years, then we can expect exabytes (millions of terabytes) of data per day. According to various estimates, from hundreds of thousands to tens of millions of self-driving cars may appear on the roads by this time.

5G as the Key to Success

As mentioned above, self-driving cars can receive information about pedestrians and cyclists not only from their sensors but also through the exchange of data with other cars, traffic lights and other urban infrastructure.

Several 5G connected car projects already exist. Cars use the mobile carrier’s 5G network and C-V2X (Cellular Vehicle-to-Everything) technology to communicate with other cars, cyclists, and even traffic lights. The latter are equipped with thermal imagers that detect pedestrians approaching the crossing; as a result, a warning appears on the car’s dashboard. Connected cyclists are informed of their location, which prevents dangerous situations. In case of poor visibility, the parked cars automatically turn on the emergency flasher lights, notifying all approaching cars of their position.

The capabilities of 5G mobile networks come in handy here. They provide fast speeds, very low latency, and the ability to support a large number of simultaneous connections. Self-driving cars without such data processing capabilities will not be able to perform many tasks faster than a person. For example, to determine the appearance of a pedestrian at the nearest crossing. Moreover, delays should be minimal, since even a fraction of a second delay can lead to an accident.

Major car manufacturers such as BMW, Daimler, Hyundai, Ford, and Toyota are already integrating 5G technology into their products. Billions of dollars have already been spent by cellular operators building 5G networks. So, this is the right time to give vehicles a set of skills that will be useful in everyday operation.

All experiments with 5G-connected self-driving cars will come to a standstill unless a 5G infrastructure is in place. Again, an unmanned vehicle can generate 1TB of data per hour, so the mobile network must be ready to transfer this data.

How to Process and Store Exabytes of Data

Not all data types require immediate processing, and the onboard computer has limited performance and storage capabilities. Therefore, data that can “wait” should be accumulated and analyzed in peripheral data centers, while some of the data will migrate to the cloud and be processed there.

It is the responsibility of city governments and automakers to capture, process, transfer, protect and analyze data about every car, traffic jam, pedestrian, or pothole. Some smart city architects are already experimenting with machine learning algorithms that analyze traffic data more efficiently to quickly identify potholes in the road, regulate traffic, and instantly respond to accidents. From a global perspective, machine learning algorithms provide recommendations for improving urban infrastructure.

To introduce fully autonomous driving into our life, it is necessary to solve the problem of processing and storing vast amounts of data. Every day, an unmanned vehicle can generate up to 20 TB of data. Just one car! In the future, it can lead to exabytes of data being generated in one day. To store this data, you need a high-performance, flexible, secure, and reliable edge infrastructure. There is also the problem of efficient data processing.

For the onboard computer to make real-time decisions, it needs the most up-to-date information about the environment. Old data, such as information about the location of the car and speed one hour ago, is usually no longer needed. However, this data is useful for further improvement of autonomous driving algorithms.

Developers of artificial intelligence systems must receive large amounts of data in order to train deep learning networks: identify objects and their movement through cameras, lidar information, and optimally combine information about the environment and infrastructure in order to make decisions. For road safety specialists, the data collected by cars immediately before accidents or dangerous situations on the road is vital.

As data is collected by self-driving cars and transferred from them to peripheral data centers, after which it migrates to cloud storage, the issue of using an optimized and tiered data storage architecture becomes more and more relevant. Fresh data must be analyzed immediately to improve machine learning models. High throughput and low latency are required here. SSDs and high-capacity HAMR drives with support for multi-drive technologies are best suited for this purpose.

After the data has passed the initial analysis stage, it must be stored more efficiently: on high capacity but low-cost traditional nearline storage. These storage servers are well suited if the data may be required in the future. Old data that is unlikely to be needed, but must be kept for some other reason, can be moved to the archiving level.

Data will increasingly be processed and analyzed at the edge, ushering in the era of Industry 4.0, which is changing how we use data. Edge computing will allow data to be processed close to where it is being collected, rather than a traditional cloud server, allowing it to be analyzed much faster, immediately responding to changing situations. A high-speed network of information exchange between cars and peripheral data centers will help make autonomous driving safer and more reliable.

Conclusion

Hopefully this analysis has shed some light on how important data is in the field of autonomous driving. Mass adoption of unmanned vehicles involves the collection of plenty amounts of data that should be processed not only by the onboard computer but also by edge servers and the cloud. The data processing infrastructure should be ready beforehand.

As the adoption of 5G spreads, self-driving cars will begin to generate more and more data, which will then be analyzed and used to make smart cities a reality. Achieving this goal will not be very easy, but in the end, we will open a new chapter in the history of such a popular means of transportation as a car.

Self-driving cars are at the forefront of artificial intelligence technologies, communications, and data storage. To reach the level of fully autonomous driving, it is necessary to continue the development and improvement of these technologies.