Autonomous vehicles have dominated the news for much of the past decade, but as 2021 looms, we seem to be no closer to being whisked from location to location without needing to be stationed behind the wheel. Here’s a look into why big data is setting itself up to be an even bigger hurdle for the future of driving.
Tesla caused a stir in recent weeks with the announcement that it was running beta tests on its ‘Full Self-Driving’ software. The news led to many enthusiasts letting their imaginations run wild at the prospect of cars capable of driving themselves from A to B, but in reality, the innovative manufacturers haven’t quite managed to produce software that lives up to its ‘full self-driving tag, with Tesla stating that the mode still requires ‘active driver supervision’ to protect the safety of other road users and pedestrians.
(Image: Ptolemus)
As the graph above shows, despite Tesla’s efforts, we’re unlikely to see high-automation driverless cars until later in the 2020s, with adoption for this form of vehicle to be minuscule until the 2030s at least.
The slow rollout of autonomous vehicles may be surprising for some, given how frequently the likes of Google, Uber and Tesla have been testing self-driving cars, but the reason why we may not be seeing truly self-driving vehicles on the roads for some time may lie in the huge levels of big data that the cars will need to process in order to successfully interact with its surroundings without mistake.
Catching Up With Human Intelligence
Although the technology driving autonomous vehicles is developing at a rapid rate, roboticist and AI researcher, Rodney Brooks, believes that driverless cars will need to display a level of intelligence akin to humans in order to understand and act efficiently in the face of the many ‘edge-cases’ that drivers encounter on a regular basis.
“Even with an appropriate set of guiding principles, there are going to be a lot of perceptual challenges that are way beyond those that current developers have solved with deep learning networks, and perhaps a lot more automated reasoning than any AI systems have so far been expected to demonstrate. II suspect that to get this right we will end up wanting our cars to be as intelligent as a human, in order to handle all the edge cases appropriately,” Brooks wrote.
To practise safety on roads across the world, self-driving cars will not only need to be capable of driving, accelerating and braking at the right times, but they will also have to process huge volumes of information about their respective environments, road signs, traffic markings and other drivers in real-time to ensure the safety of everybody around them. This level of understanding can be learned in order to become intuitive among humans, but with artificial intelligence, having to interpret so many signifiers can be largely problematic when the stakes are so high.
Getting Machine Learning Right
Autonomous vehicles typically turn to AI and machine learning in order to process the big data that arrives from a car’s various sensors. This helps to interpret the scores of data quickly so that decisions regarding actions can be made in real-time.
The algorithms that work to identify objects detected by sensors and classify them correctly as pedestrians, other cars, street lights and so on are important factors in ensuring that cars safely navigate their environments. Subsequently, the cars interpret the information to quickly decide whether affirmative action needs to be taken, such as fast-braking or swerving.
(Graph showing current and anticipated machine learning maturity in companies during 2020. Image: Statista)
Over the coming years, the machine learning available to detect and classify objects and obstacles from various data points will develop to a level of awareness and observation that beats the field of vision and responsiveness of human drivers. But for now, there’s no agreed basis for ensuring that machine learning algorithms used in autonomous vehicles are safe. Subsequently, there’s no all-encompassing agreement across the industry on how machine learning should be trained, tested, or validated for self-driving cars.
The development of a machine learning algorithm that can actively decipher the scores of data that are presented to an autonomous vehicle and safely guide a car from A to B is essential for the industry, but it may take a significant amount of time before a working model wins the approval of regulatory bodies.
The Price of Technology
There’s little doubt that in the future, we’ll see a coming together of technology advanced enough to be able to decipher the constant stream of data covering the world around it while turning it into safe and efficient actions in real-time.
We’ll also eventually see cars capable of interacting with one another using IoT technology that interconnects every vehicle on the roads to ensure crashes don’t take place. However, the costs associated with such highly advanced technology today are set to be astronomical for manufacturers.
While the first manufacturer to develop a fully functional autonomous vehicle that’s approved by regulators may represent something of an industry holy grail, the cost of condensing such advanced technology into a self-driving car is likely to send the price of cars soaring even after they’re given the green light to hit the roads.
It may take a giant initiative for manufacturers to encourage autonomous vehicle purchases in the future. However, with many nations pledging to end the sale of petrol cars over the next five to 20 years, governments around the world may eventually create schemes to subsidise the costs of electric autonomous vehicles in the future.
However, back in the present day, the world is still captivated by the boundless potential of autonomous vehicles, although we may have to face the prospect of being captivated by these thoughts for many years to come.
Top comments (0)