Nvidia: Autonomous vehicles require time and powerful technology
See Full Size
On the other hand, Ali Kani, the head of the company’s automotive unit, thinks that we are still not close to fully autonomous vehicles. According to Kani, speaking to Autocar, it does not seem possible to achieve fully autonomous driving in the 2020s. Arguing that we are not close to real autonomous vehicles, the manager said, “This is the subject of the 2030s. We are not close yet. This is a really difficult task.”
Kani noted that although a few vehicles offer limited autonomous capabilities in certain controlled situations, a significant advance in processing power and technology is required for vehicles to be fully self-driving. “The software we are developing now is completely different from the software we developed last year,” said Kani, and continued:
“We are now working on large language models like ChatGPT that work with video, and three years ago there was no one in the automotive industry doing this. Such models require much more processing power and much more memory bandwidth. You need more sensors, such as lidar and radar, and redundant algorithms are required to ensure its security. “All of these must run in parallel, which means more processing power.”
See Full Size
“When the vehicle follows the planned rules, you experience choppy movements and unnecessary braking. As a result, ‘I don’t like this car, I don’t feel safe.’ you say. However, the next generation of vehicles will learn behaviors, so natural understanding will emerge. That’s when you think, ‘Wow, this car drives so calmly and smoothly.’ you will say.
Kani also underlined that caution should be exercised in the development of autonomous driving technology.
“The industry needs to act slowly on this issue. If one company makes a mistake, the entire industry goes back several years. Therefore, we must act in the most responsible way and not take any shortcuts. “Once you prove that it is truly safe, you can do it.” he said.
This news our mobile application Download using
You can read it whenever you want (even offline):
