Waymo and Uber offer new AI technology for robomobiles

Waymo and Uber offer new AI technology for robomobiles

June 23, 2020 0 By autotimesnews

At the last online conference on computer vision CVPR 2020, specialists from Waymo and Uber presented their developments to improve the reliability and safety of autonomous vehicle control systems. Waymo described the ViDAR framework, and Uber demonstrated a vehicle-to-vehicle communication technology.

Two reports of companies competing among themselves in the field of autonomous transport were presented at the CVPR 2020 conference, which was attended by machine vision system developers from more than 30 countries, writes Venture Beat. Drago Anguelov spoke on behalf of Waymo and Google Brain about the ViDAR system, and Raquel Urtasun, head of Advanced Technologies Group Uber, demonstrated V2VNet technology.

Vidar

The ViDAR project, a collaborative effort by Waymo and Google Brain, uses machine vision to isolate structure from movement. The system learns three-dimensional geometry from a sequence of images or frames shot with cameras mounted on a car. To do this, she uses parallax, a change in the apparent position of the object relative to the distant background, depending on the position of the observer. With a couple of pictures and lidar data, ViDAR can predict future camera positions and image depth.

ViDAR is used on Waymo robotic cars to provide high-quality camera-oriented depths, evaluate camera movement relative to the scene and build dynamic models. The solution creates a model that predicts the direction of movement of obstacles, in particular pedestrians.

V2vnet

Uber experts have developed the V2VNet system, which provides autonomous vehicles the ability to efficiently exchange data over the air. With its help, machines connected to the network can send data to each other, giving more complete information about the traffic situation. The system also allows you to select only relevant data from the information stream.

The results of several experiments showed that the error level of V2VNet is lower by 68% compared to an individual robomobile.

It is not yet clear whether V2VNet will be used in practice, but Waymo, a Uber competitor, is already using wireless data exchange among its Chrysler Pacifica unmanned minivans, which tell each other about road hazards and route changes.

This spring, Waymo introduced the fifth generation of unmanned control systems. Compared to previous versions, the fifth radar has increased resolution, coverage and field of view.