Just when you think that self-driving cars will be the solution to the unacceptably high rate of road traffic accidents across the globe, you read about accidents such as the 2018 fatal Uber crash in Tempe Arizona – thought to be the first death ever caused by a self-driving car. The incident, reported investigators from the National Transportation Safety Board, was the fault of the driver – who was busy watching a video on her smartphone contrary to the manufacturers’ safety policy. In this case, the car rammed into a pedestrian, the system identifying the victim as a bicycle moving into the vehicle’s path. The identification was made 1.2 seconds before the fatality – too late to prevent the tragedy from occurring. Uber now claims that it is now relying on a far more sophisticated artificial intelligence (AI) technology that will enable it to improve predictions. What are the basis of these claims, and how long will it be before autonomous cars are safer than human drivers?
Uber’s New SC-GAN
The groundbreaking technology is called SC-GAN (or scene-compliant GAN). Its critical feature is a differential rasterizer, which permits generated trajectories to be projected directly into the raster space in a differentiable manner. SC-GAN essentially creates trajectories that follow constraints existing within scenes. The detection and tracking systems receive information from radar, lidar, and the car’s x and y axes. It enables cars to be 10 times more precise in predicting traffic movement. SC-GAN reduces ‘false positives’ by an impressive 50% compared with baseline measurements.
How Our Idea Of Dependability Will Change
For the average vehicle buyer, dependability is key in vehicle selection. Research compiled by CEA Insurers shows that reliability is right up there with price and safety when it comes to top considerations prior to purchase. Today, reliability involves having few or no issues with engine electrical systems, reliable suspension, optimal steering and sat-nav, and the like. In the future, AI will change driver’s expectations, with dependability being a given. This technology will be used to monitor dozens of sensors, alerting passengers not only to dangerous situations, but also to maintenance and repair needs. Dependability, then, will take the back seat, and safety and comfort will most likely be the prime considerations for buyers of new vehicles.
How Close Is An AI-Driven Future?
AI technology is constantly being refined, but experts predict that it will at least 12 years before fully autonomous vehicles are sold to private buyers. Even robotic taxis or ubers will not will be ready until at least 2025. By 2034, says a J.D. Power survey, only 10% of all vehicles bought will be autonomous. One reason for the delay is the current demand for electric vehicles and the general fear that AI is still not sophisticated enough to make sharing the road with this type of vehicle (or riding in one) fully safe.
Reaching Level 5
Developments such as Uber’s SC-GAN are bringing the world a little closer to full autonomy, but the latter is still worlds away. There are five levels of autonomy: Level 1 is where some cars are already at now (e.g. their car may have lane keeping or automated braking functions). Level 2 combines two or more sophisticated autonomous driver functions. At Level 3, cars can drive from point A to B without human intervention, but only under certain conditions. At Level 4, the driver can fully relax if the car is driving within a geofence or if other restrictions are present. It isn’t until a vehicle reaches Level 5 that it can drive fully autonomously in any situation, without the help of the driver. The SC-Gan makes Level 3 driving a little safer, but it arguably still fails to position a vehicle at Level 4 safety.
Uber’s new SC-GAN represents a major development in AI technology for self-driving vehicles. The system enables vehicles to predict occurrences on the road with much greater accuracy. It is just one of many new developments taking place that reveal that safety continues to be the number one stumbling block to cars having a great autonomous capacity.