Earlier thought of as science fiction, self-driving cars are a reality today. Thanks to the integration of several various automotive technologies that have matured over the years. These technologies include ultrasonic sensors, bumper mounted radar, V2V/V2I communications, software to interpret road signs, cruise control, and much more. Self-driving vehicles promise to increase accessibility for those who are unable to drive including the young, elderly, and the mobility impaired individuals. Autonomous vehicles are also expected to decrease the demand for parking, allowing some parking infrastructure to be used for a different purpose. Furthermore, as autonomous vehicles are not dependent on a human’s range of vision and can communicate with each other, they are likely to enhance the safety of vehicles.
While companies claim that self-driving vehicles will make our trips more convenient, people are still skeptical about the technology used in the cars. Their biggest concern is safety. There have been several accidents involving self-driving vehicles that took place recently. One of them is the crash of Waymo’s autonomous minivan on the highway in California in October 2018. The car was run by an autonomous software. However, the human driver fell asleep and accidentally pressed the gas pedal, which turned off the autonomous mode, eventually causing the accident. In March 2018, a self-driving car operated by Uber killed a woman on a street in Arizona. The vehicle was equipped with special sensor tech capable of identifying obstacles but failed to detect the pedestrian and struck her. A safety driver in the car didn’t have her hands on the steering wheel.
Car manufacturers and researchers are actively involved in testing their autonomous vehicles to make them safe in every possible way. For instance, the Stanford scientists discovered a technology to be used in self-driving cars which enables them to learn from previous experiences, thus helping the vehicles to drive more safely in unpredictable and dangerous situations. The technology was tested on two self-driving vehicles namely, Niki and Shelley on a racetrack. It performed as well as a racecar driver and an autonomous control system.
While autonomous cars that exist now may depend on in-the-moment evaluations of their environment. However, the control system developed by the researchers includes data from recent maneuvers and driving experiences of the past. Such control systems need to access information about the road-tire friction, which tells about the braking capacity of cars, acceleration, and steers to stay on the road in emergency cases.
If an autonomous car is to be safely operated to its limits, it must be fed with details such as road-tire friction in advance. This is difficult as friction is variable. To create a more responsive control system, the scientists developed a neural network that incorporates data from past driving experiences and a winter test facility with foundational knowledge provided by 200,000 physics-based trajectories.
The group performed piloting for their new system. Shelley performed by the physics-based autonomous system, fed with data about the roads and conditions. When compared on the same road during 10 trials, Shelley and a skilled driver created comparable lap times. Niki was loaded with the new neural network system. The car performed in the same manner, running the learned and physics-based systems even when the neural network lacked information about road friction. In simulated tests, the neural network system performed better than the physics-based system in high-friction as well as low-friction situations. It ran especially well in conditions that combined the two situations. The scientists say that though the results are encouraging, their neural network fails to run efficiently in conditions other than the ones experienced. They said that as the cars generate more data to train their network, they should be capable of performing more tasks.