Once again, Google’s sister Waymo takes a look behind the scenes and reveals that a new architecture of artificial intelligence.
That the Artificial intelligence is becoming increasingly important in autonomous driving is clear. Also at Waymo one puts with preference on this technology.
AutoML scheme. Source: Medium.com
Especially in the field of sensor evaluation and the intention calculation Other road users, the artificial intelligence helps the automated vehicle. It uses a combination of different neural networks for sensor evaluation, object identification and tracking of movements. But training and fine-tuning this technology takes a lot of time.
That’s why they’ve teamed up with the Google Artificial Intelligence Department, the Brain Team. The goal is to automatically train artificial intelligence. In addition, the team is equipped with more modern systems. That’s the way to work with AutoML, where Auto stands for Automatic and Machine Learning ML.
AutoML technology is delivered in high volumes and continuously. In addition, the artificial intelligence are faster in the calculation, so have a lower latency or also called inference time. This means that the result is faster, which is enormously important as the speed increases. Specifically, you can achieve a latency of 10 ms. This development should make it possible to realize new environments faster.
In the beginning, one pursued strategies for improved object recognition. For this purpose, an automatic search algorithm has been developed which Lidarbilder analyzed. But one had to decide whether one wanted to achieve a higher quality or a lower latency. Then this concept was applied to the recognition of lane markings and was successful.
Therefore, this concept has evolved to limit search selection, resulting in high quality with low latency. The training was also shortened – from months to hours. The concept at Waymo is called the proxy end-to-end search, which has been applied to lidar systems.
The result was a reduction in latency of 20 to 30 percent with the same quality. With higher quality, we achieved a roughly ten percent reduction in the error rate, with the same latency. This development has been integrated into the vehicles. Nevertheless, they are working on improving this AI architecture AutoML and want to apply it to other areas as well.
source (English)
About David Flora
I’ve been writing about the topic of Autonomous & Connected Driving since 2011 and also talk about it on other sites like the Smart Mobility Hub. I studied social sciences at the HU Berlin and since 2012 I am a freelance journalist. Contact: mail@autonomes-fahren.de