A fully autonomous vehicle uses an automatic driving system that doesn\u2019t need a human driver. Most large automotive manufactures are in the race to develop autonomous vehicles, alongside newer companies such as Google, Tesla and Uber. But how close are they to creating the technologies for a truly self-driving vehicle?<\/p>\n
Developments in autonomous vehicle technology often stem from the development of connected cars. Vehicles are increasingly able to communicate with internal and external systems, which is a necessary step for autonomy. For instance, more and more models now have the ability to process the environment around them with parking assistance and lane control. These developments are becoming the building blocks for autonomous cars.<\/p>\n
Vehicle-to-vehicle (V2V) communication is potentially the next stage in building the technology for autonomous cars. It enables connected vehicles to exchange information though a wireless, ad hoc network about factors such as position, speed, direction, braking and acceleration. Proponents of the technology claim that it can help prevent accidents, reduce carbon emissions and ease congestion.<\/p>\n
A report by Juniper Research<\/a> suggests that, in the US, 775 million consumer vehicles will be connected by 2023, rising from 330 million vehicles in 2018. At present, however, many initiatives are on hold as manufacturers wait for the standardisation of technology. Some expect dedicated short-range communications (DRSC) to be the standard, but the advent of 5G is making other manufacturers sceptical of the long-term viability of DRSC.<\/p>\n V2V communication will become part of the autonomous car ecosystem. This larger ecosystem is called vehicle-to-everything (V2X). Cars will be able to continually communicate with a whole range of items, from other cars to pedestrians\u2019 smartphones to items in the transport infrastructure, such as traffic lights. At present, DRSC is the main standard for these developments. The US National Highway Traffic Safety Administration, the International Standards Organisation and the European Parliament have all backed DRSC as the standard for V2V. Volkswagen have opted for WLANp, the technology behind DRSC, in the Car2X system in their new Golf 8<\/a>.<\/p>\n The sheer number of VW Golfs on the road, plus Volkswagen\u2019s plan to roll out the technology to other vehicles, means that the technology could become increasingly widely used. This in turn may encourage other manufacturers to adopt DRSC. However, there is concern that it may be too slow for very fast-moving vehicles, so lower-latency 5G may eventually be preferred by some manufacturers, leading to a hybrid approach using alternative technologies.<\/p>\n As autonomous communication builds to V2X, vehicles will increasingly need reliable networks. These will require redundant architectures to increase reliability, operating in real-time.<\/p>\n V2X will not only include V2V, but also V2I (infrastructure) and V2N (networks such as infotainment systems). Vehicles will also handle their own high-performance systems of sensors, computers and cameras. This will generate a huge amount of data.<\/p>\n A report by Hitachi<\/a> estimates that fully autonomous cars will send 25 gigabytes of data to the cloud every hour. According to Tesla\u2019s Elon Musk, \u201cIt\u2019s actually quite a challenge to process that data, and then train against that data, and have the vehicle learn effectively from the data, because it\u2019s just a vast quantity.\u201d<\/p>\n Vehicle manufacturers will need to constantly increase bandwidth to manage the flow of data. This will require sophisticated electronics inside the vehicle to create a high-speed information highway. An ethernet working group at internal standards association IEEE has endorsed 1000BASE-T1 Gigabit ethernet as the new network bandwidth standard for the automotive industry.<\/p>\n V2X communications must continuously capture data from the surrounding environment. Most manufacturers are using combinations of radar, ultrasonic (which senses distances using ultrasound), video cameras and lidar (which measures distances using light in the form of a pulsed laser).<\/p>\n However, in trials automonous vehicles have struggled to interpret their environment using these types of sensor. In particular, heavy traffic and certain weather conditions have been problematic. For instance, glare from the sun can affect what video cameras can see. Lidar is limited by distance as are ultrasonic sensors, and is compromised in heavy rain.<\/p>\n Companies are now incorporating other technologies, some of which are derived from military applications. Sensor manufacturer AdaSky has demonstrated how its thermal sensors can detect people at a distance, even in heavy rain. Radar company WaveSense has deployed ground-penetrating radar that uses the geological features below the road to determine a vehicle\u2019s location, which also has the benefit of removing the issues created by human-made changes above ground.<\/p>\n Machine learning, where machines have access to data to continuously learn and evolve, will be an important factor of AI in autonomous vehicles. The sheer scale of variables involved when a vehicle drives itself means that autonomous cars will have to continuously adjust their knowledge base.<\/p>\n Vehicles will be required to interpret sensor data and to then make judgements about it. Autonomous driving requires the vehicle to not just perceive a person pushing a buggy, but to know how that person will behave and how the vehicle should respond.<\/p>\n Some vehicles already use machine learning for Level 3 autonomous driving, which is where the vehicle is partially automated under certain driving conditions but still has a human driver. However, for Level 5 (full automation) to become a reality, major advancements in machine learning are required. This has led to a range of investments by large motor companies, such as Ford\u2019s billion dollar investment in start-up Argo AI. In 2019, Argo AI funded an Autonomous Vehicle Research Center at Carnegie Mellon University<\/a>, an institution world-renowned for its machine learning research.<\/p>\n Given the complexity of autonomous vehicles, we remain years away from riding in fully self-driving cars. General Motors, Ford and Renault-Nissan have all pushed back<\/a> on initial deadlines based on or around 2020.<\/p>\nData and systems<\/h3>\n
Sensors<\/h3>\n
Artificial Intelligence (AI)<\/h3>\n
Autonomy remains very much a work-in-progress<\/h3>\n