Imagine yourself sitting in a car – hands no more on the steering wheel, feet no more on the accelerator. Imagine speaking to your car, telling it where to you, giving it some instructions, and then sipping on coffee while it drives you to your destination – just like a science fiction movie.
Similar must have been the imagination of Sebastian Thurn, an associate professor of computer science and director of Stanford Artificial Intelligence Laboratory who won the Urban Challenge Competition by Defense Advanced Research Projects Agency (DARPA) in 2005 with his autonomous robotic car ‘Stanley’. The car was equipped with multiple sensors and backed with custom-written software, including machine learning algorithms that helped the vehicle in finding paths, detecting obstacles, and avoiding them while staying on the course.
The concept of autonomous cars was first exhibited in 1939 at the Futurama exhibit at New York World’s Fair. The exhibit was designed to give people an idea of how things would look like 20 years in the future. However, it was only after Sebastian Thurn’s project that self-driven cars actually inched closer to the roads.
In 2009, he led the self-driven car project of Google with the goal of autonomously driving over 10 uninterrupted 100-mile routes. By 2010, the advances made by the project were visible to all the major automakers, including Apple and they started investing in the development of their own autonomous cars. The next decade saw billions of dollars-worth investment in the industry.
In 2016, an autonomous driving technology company called Waymo became a subsidiary of Alphabet, and Google’s autonomous car project ultimately became Waymo. Since then, Waymo has been leveraging AI extensively to make fully autonomous cars a reality.
“While perception is the most mature area for deep learning, we also use deep nets (DNN) for everything from prediction to planning to mapping and simulation. With machine learning, we can navigate nuanced and difficult situations; maneuvering construction zones, yielding to emergency vehicles, and giving room to cars that are parallel parking,” Dmitri Dolgov, CTO and vice president of Engineering at Waymo wrote in a blog.
The team led by Thurn has extensively trained Waymo’s deep learning modules for over 10 million miles on roads and hundreds of millions of interactions between pedestrians, cyclists, and vehicles. It has also covered over 10 billion miles in simulation. Waymo’s cars use high-resolution cameras and lidar technology (a way of estimating the distance to other objects by bouncing light and sound off things). It helps them identify other vehicles, pedestrians, and obstacles while moving.
Waymo has announced in 2018 that their self-driven cars will arrive on roads by 2020. However, that has not yet happened. The available technology is still limited to cars that can automatically brake for you in case they detect a collision, keep you in your lane, and handle highway driving. However, a lot of work still needs to be done to produce a vehicle that can complete a journey legally and safely without human intervention.
Three key barriers that stand on their way to the market are technology, regulations, and public acceptance.
Technology – While AI algorithms can detect and classify objects accurately, they can’t mimic the intricate complexities involved in driving. The car not only needs to recognize humans and other objects but also know what to do in unfamiliar circumstances.
Regulations – Policymakers and regulators have not yet come up with the criteria to determine whether self-driven cars are safe for open roads. A graduated certification approach is needed where the autonomous vehicles will first be evaluated in simulations and then in a controlled real-world space. Only after that, they can be run on open roads.
Public Acceptance – The people should be willing to accept and accommodate autonomous vehicles on the roads and be benevolent towards them. A lack of trust and cooperation will not only affect the safety of those using such vehicles but also those who share the road with them.
Unless the automakers and regulators establish a safe and guided environment for such vehicles, the concept of completely autonomous cars will take decades or even more to become a reality.
Source: indiaai.gov.in
