Enabling The Self-Driving Car

The "software-ization" of telecommunications infrastructure is an example of the technological and economic maturity of a number of trends, such as the diffusion of ultra-broadband fixed and mobile networks, the increased performance of IT systems combined with costs reductions, and development of increasingly powerful end devices, including cars with embedded information communications (ICT) technology.

With the shift to software-based network and service functions, logical resources such as virtual machines are being decoupled from the underlying physical infrastructure, which increases flexibility and automates operations. At the same time, it’s possible to optimize these new architectures to potentially reduce end-to-end latency dramatically to a level that’s not yet possible, which opens up the potential for new service paradigms such as cognition-as-a-service, as well as new ecosystems.

An example such of an ecosystem involves the ability to actuate, through the network, remote control of mobile robotic systems with complex cognitive capabilities and characteristics of autonomy that could operate in an unstructured and very dynamic environment. Truly mobile robots, remotely controlled with ultra-low latencies, could have an enormous impact on industry, agriculture and several social applications, such as the home and smart cities.   

This could lead to an application often depicted in science fiction: self-driving cars . Ultimately, a self-driving car is a complex robotic system equipped with sensors, actuators and ICT capabilities. Driving a car in real traffic is a challenging task for machine intelligence: unit of milliseconds of reaction time are required to avoid sudden and unpredictable obstacles, and inherent “common sense” also is necessary. These characteristics mean a self-driving vehicle would require a lot of computing power to minimize application latencies, as well as very low network latencies.

Today, the local computing power equipped in a self-driving car is not enough, due to space, heat dissipation, and the cost of executing the heuristics or artificial intelligence needed to provide such levels of autonomy. However, we have enormous computing and storage power in cloud infrastructures, so why not execute the necessary intelligence in the cloud to drive a car? This can be seen as simply another facet of cloud robotics.

But there is a problem: A primary bottleneck is the overall sum of application and network latencies, which are far too high today. The software-ization of telecommunications infrastructures, however, will make it possible to overcome this problem. Ultra-broadband radio connectivity in the form of 5G that is expected to arrive by 2020, coupled with the flexibility of orchestrating huge amounts of computing and storage power -- from centralized clouds to local, distributed mini-data centers closer to cars -- will allow us to execute this machine intelligence.

Automated network operations will allocate and control cognitive tasks to trigger actions with ultra-low latencies based on the sensed environment, making intelligent mobile robots and  self-driving cars a reality.