Apple wants to use machine learning on the Apple Car because current processors aren’t fast enough to independently make key decisions while driving.
The fact that Apple wanted to take advantage of machine learning on the Apple Car already seemed clear when the company chose John Giannandrea, the company’s head of artificial intelligence, to lead the autonomous driving project. Now, a new patent confirms these intentions.
The system is linked to the fact that decisions made at the wheel must be extremely fast. Even a correct decision, for example on changing lanes or avoiding a collision, can be fatal if not taken quickly. “Until relatively recently”indicates the patent, “Due to the limitations of available hardware and software, the maximum speed at which calculations could be performed for the analysis of relevant aspects of the vehicle’s external environment was insufficient to enable navigational decisions without human intervention”.
However, even current processors and those expected in the coming years may not be enough: “Even with today’s fast processors, large memories and advanced algorithms, the task of making timely and sensible decisions about the vehicle environment remains a significant challenge. »
The patent demonstrates the complexity of the decision-making process when driving autonomously, as cars will never drive alone considering that there will always be unpredictable behavior from drivers of other cars. Additionally, the real world is far more crowded than any test environment, so Apple also notes that self-driving decisions are currently being made in the presence of incomplete data.
In over 17,000 words, the patent describes situations related to the “action space” of the car. This is the time and distance in which the car must make its decisions. “In some states, such as when the vehicle is traveling on a largely empty straight highway with no ability to turn for several miles, the number of actions to be assessed may be relatively low; in other situations, such as when the vehicle is approaching a crowded intersection, the number of actions could be much higher”.
In any case, the car’s systems must determine the current state of the environment around the vehicle. For this reason, it may be necessary to identify “a corresponding set of feasible or proposed actions that can be taken”. An action can be “turn left” or “change lanes”. At least in some cases, machine learning can be used to help the car assign a number or value to each possible decision and thus determine the best course of action.
Will this be the heart of Apple’s autonomous driving?