Solving Autonomous Car Problems: Are There Any?

A few years ago, the future of autonomous vehicles appeared more than controversial. Announcements of major auto manufacturers and software pioneers were overshadowed by countless reports of autonomous car problems.

Call me crazy, but all those driverless car incidents prove the great potential of the technology. Why? Because mistakes are the foremost mark that proves the process is going.

Today, the majority of problems with driverless cars have been resolved by the common efforts of car manufacturers and software vendors. The trend is clear: overcoming the remaining issues is only a matter of time. So, get prepared for the autonomous vehicles.

The statistics of autonomous car problems give reason for optimism

Each year, Original Equipment Manufacturers (OEMs) of self-driving vehicles have to reveal statistics on the number of problems with  autonomous cars: incidents and failures. The most recent report appears ominous at first sight. According to the data, 2,244 self-driving fails occurred in 2017 on the roads of California. Fortunately, people were hurt in only a few cases.

Notably, the available reports don’t include the number of disengagements that occurred with Tesla cars. This is because Tesla is the only company in the world whose autopilot system is being tested primarily by end users in so-called shadow mode. These real-world driving tests are not regulated by either federal or California law.

Source: Department of Motor Vehicles, State of California

Referencing self-driving incidents and problems with autonomous cars as well as statistics from the past may help us grasp the real trends underlying the technology’s development. Over the last year, the number of safety-related incidents on California roads has fallen by almost 300 disengagements. This improvement is remarkable. It shows how affordable and secure solutions of problems with self-driving vehicles are being brought to life.

Heavy weather conditions? Causing problems with autonomous vehicles?

Until recently, one of the main challenges holding back self-driving technology was its inability to perform well in rain and snow. Autonomous vehicles identified precipitation as a solid object that had unexpectedly dropped in the way. As a result, they braked and stopped.

In cooperation with the University of Michigan, Ford was the first to present a comprehensive solution to this issue. They improved their system of LIDAR sensors so the laser bursts not only could create a 3D map of the environment but also could figure out whether they were hitting raindrops or snowflakes. To put it simply, improvements to the algorithm were enough to resolve problems with autonomous vehicles under heavy weather conditions.

Driverless vehicles are getting better at communicating with people

Other problems facing autonomous vehicles that used to cause headaches for developers was communication between self-driving cars and people on the road – primarily pedestrians and cyclists but also law enforcement and highway safety employees. The question was how software could help to substitute interpersonal human contact on the road.

The solution presented by startup Drive.ai illustrates how the challenges of human-machine interactions can be met effectively. Drive.ai proposed communicating with visual displays that go beyond today’s turn signals. Namely, banner-like text and easily identifiable sounds were used to resolve this problem facing autonomous vehicles: self-driving cars communicating with humans.

Navigation is no longer a challenge

At the beginning of the 2010s, when today’s industry leaders revealed their first fully autonomous driving solutions, the issue of navigation was a huge problem. Early models of scanners and navigators weren’t good enough to allow cars to operate smoothly. But Google wasn’t going to accept this situation. The company developed a kind of Street View mode for cars in the form of exceptionally detailed virtual city maps that could be filled with objects marked by car sensors. Today, error-free navigation of driverless cars is taken for granted.

In the early 2000s, autonomous vehicles were rather an impracticable concept. Only market leaders and major car manufacturers were able to allocate the funds required to develop sophisticated software and hardware. Today, the costs of on-vehicle sensors have been reduced by advanced technologies like Convolution Neural Networks (CNNs). As a result, small producers and local startups may enter the industry on equal terms with the most influential players.

The next step? Make the emergency brakes work better

You’ve probably heard about the 2016 Tesla autopilot accident that resulted in a person’s death. Although the accident was the result of a combination of human error and technological shortcomings, it’s clear that developers still have a lot of work to do. A more recent Tesla car incident only proves that.

One issue that remains unsolved is the limited usability of the Automatic Emergency Braking (AEB) system. If the AEB system is turned on, it’s likely that the brakes will be activated almost every time an object ahead slows down unexpectedly. On the other hand, if the system is off it means that we’re still far from full vehicle automation. Advanced computer vision reinforced by innovative use of stereo sensors is what will help autonomous cars distinguish whether AEB should be applied. In combination with deep learning provided by neural networks, stereo sensors can make fully autonomous driving safer and eliminate autonomous car problems.

Why should self-driving technology become more secure?

The future of self-driving cars is also challenged by substantial security issues. In a few years, automated vehicles are likely to become the top contributors to the Internet of Things (IoT). This means that data generated from automated cars will also become more exposed to risks. For example, in August 2017, hackers managed to steal more than 100 cars in Texas by using a computer to unlock cars and start them.

Software developers are eager to elaborate complex real-time alert systems that could use information provided by OEMs so road police could be alerted each time an automated car shows unauthorized behavior. In other words, leveraging the software security infrastructure is both a challenge and a prerequisite to a secure self-driving future and less autonomous car problems.

Building trust on the road is essential

The progress made in human-machine interactions in the last decade is remarkable. But let’s be realistic. Until interfaces of self-driving cars become more human, people will hardly feel comfortable sharing the streets with them. One of the most urgent tasks of OEMs and developers is to make self-driving cars understand how people hold themselves and move. To put it simply, driverless cars need to learn how to read human body language. Eventually, with the help of approachable interfaces, cars will be able to interact with people on genuinely trusting and safe terms.

The remarkable development of the autonomous vehicle industry in the last decade is primarily the result of close cooperation between OEMs and software vendors. Advanced solutions devised by automotive software developers have already resolved numerous issues and problems with autonomous cars, including driving in bad weather, navigating smoothly around cities, and interacting with people on the road. This cooperation between OEMs and software developers has been necessary to overcome recent challenges, including imperfections with emergency brakes, security threats, and building trust between humans and machines.

Our experts are eager to help you overcome all kinds of challenges associated with self-driving technology. Contact us and get a step ahead of your rivals in the race for autonomous, affordable, and secure driving.