Autonomous vehicles
tremendously ’, implying that the future of autonomous cars lies within electric vehicles (EVs) (Stewart 2018). In addition, the onboard computer generates heat, some so much heat that they rely on water cooling facilities to regulate their temperature, such as radiators, which themselves consume more power.
Software
As discussed in the hardware section, sensor fusion becomes more unreliable when identifying smaller objects. This process is vital in the decision-making process, so manufacturers cannot afford to make an error in this phase as an incorrect object identification could lead to a potentially fatal accident. This tragic prospect forces programmers to design the system so the computer will keep trying to conceive the object/situation to come to a definite decision (whether to brake, swerve, or remain at the speed it was travelling in). The system also has the option to brake if the algorithms cannot resolve the nearby objects, however, the question is: ‘how long will it take the system to either recogn ize the objects, or to switch into a default status and brake?’ The brake reaction time is the time taken for a human to react to a situation, and then apply the brakes. From research carried out by Heikki Summala (a professor of digital humanities) in 2000, in a sudden situation, when a collision is roughly 4 seconds away, an alert driver will have an average brake reaction time of up to 1.0 second, whereas an unalert driver will have an average brake reaction time from 1.0 second to 1.3 seconds depending on the circumstance. This means that sensor fusion will have to take a fraction of a second to evaluate and come up with a definite perception of the world. Moreover, the algorithms will have to produce the optimal solution to the scene that the sensor fusion has painted, all under a second. Even though it may seem that a computer would be able to process information faster than a human brain, the world fastest supercomputer in Japan, called the Fugaku (Mount Fuji), can carry out 442 petaflops of information processing. In comparison, a human brain can carry out an equivalent of 100- 1000 petaflops (a petaflop is a measure of processing speed), more than double the processing power of the Fugaku. However, a 100-petaflop supercomputer uses roughly 15,000,000 watts (enough power to support a city of 10,000 homes), occupies an area of an American football pitch, and requires a sophisticated and expensive cooling system to manage the large amount of heat produced. Whereas the human brain uses 15 watts (the power to barely light a light bulb) and is the size of roughly 2 fists The image below illustrates how large the Fugaku supercomputer is.
Japanese supercomputer, crowned world's fastest, is fighting coronavirus. BBC 2020
Another software-related issue are the algorithms themselves. In adverse situations or conditions, fundamental algorithms and DNNs struggle to function reliably. In unmapped areas, or places where paved roads and lane-markings are non-existent, Pathfinder DNNs
23
Made with FlippingBook interactive PDF creator