Waymo, a self-driving car released by Alphabet, has gained important milestones this year, crossing 10 million actual-world miles with the driverless cars. The previous week, Alphabet has launched a commercial self-driving taxi facility – Waymo One. However, its experts are looking out for the future.
Mayank Bansal and Abhijit Ogale, its experts explained a method of AI driver training which uses labeled data, in a blog post that was posted yesterday. Waymo has been trained on millions of marked miles from professional driving situations through supervised learning.
“In recent years, the supervised training of deep neural networks using large amounts of labeled data has rapidly improved the state-of-the-art in many fields, particularly in the area of object perception and prediction, and these technologies are used extensively at Waymo,” the researchers inscribed. “Following the success of neural networks for perception, we naturally asked ourselves the question: … can we train a skilled driver using a purely supervised deep learning approach?”
The researchers developed a neural network – dubbed ChauffeurNet – in an effort to develop an AI system with the ability to mimic a professional driver. The system learned to create a driving route by seeing a combination of simulated and real data, comprising of surrounding objects, a map, the past motions of cars, and states of traffic lights. After that, a low-level controller transfers the ten-point path to acceleration and steering commands, permitting the AI system to drive both digital and real cars.
The developed model was given examples from “the equivalent of about 60 days of expert driving data,” using methods that made sure that it didn’t conclude from past motion and in fact responded to environmental changes. In experiments, it reacted to traffic controls like traffic lights and stop signs, but unsurprisingly performed badly to unknown situations.
The researchers mentioned a problem that driving demos of actual-world driving are unfair as they only are the examples of driving in good environments. To compensate, the researchers created collisions with objects and near-accidents, which stimulated the Artificial Intelligence model to prevent them.
ChauffeurNet did better in a simulated situation with the losses and keeping in account the created examples, even succeeded in shoving around parked cars, recover from little deviations in its path, and stop for a traffic light changing from yellow to red.
The researchers further mentioned in the blog post:
“Fully autonomous driving systems need to be able to handle the long tail of situations that occur in the real world. The planner that runs on Waymo vehicles today uses a combination of machine learning and explicit reasoning to continuously evaluate a large number of possibilities and make the best driving decisions in a variety of different scenarios … Therefore, the bar for a completely machine-learned system to replace the Waymo planner is incredibly high, although components from such a system can be used within the Waymo planner, or can be used to create more realistic ‘smart agents’ during simulated testing of the planner.”
