Amazon has unveiled a new version of the unmanned machine DeepRacer, which is positioned as a platform for the study of machine learning algorithms by engineers and programmers. The new model received a stereo camera and lidar. In addition, the company will add obstacles to the races, as well as allow the races of two models at the same time, according to a press release of Amazon.
In 2018, Amazon unveiled the AWS DeepRacer project, in which students and budding developers can compete in the small driverless car races. To do this, participants can use the company’s cloud service, where they first create a model and train it, and then test it on a virtual racing track. Amazon has also created a league in which developers can compete with each other in real races using the DeepRacer, which in the first version drove on a simple track, focusing on the data from a single camera.
In the new version, Amazon introduced the DeepRacer Evo with new sensors. Now it receives data about the surrounding space with the help of two cameras, as well as lidar on the roof. In early 2020, the new version will be available as a full-fledged machine, as well as as a set of sensors, which can be installed on the old version.
To take advantage of DeepRacer’s new features, developers have added new modes to the competition. Firstly, if earlier participants competed in the speed of the track during single races, now thanks to the presence of a comprehensive lidar they will be able to compete directly during the races involving two cars. Secondly, on the track, there will be obstacles that cars will need to detect and go around. In addition, the company increased the number of real races in different countries to 30 and the number of virtual to 24.
According to a similar model, the Roborace project is developing, in which the organizers provide developers with identical full-size unmanned vehicles with cameras, lidars and radars, and developers try to get around their rivals. As part of the project, the organizers created a completely unmanned vehicle with an unusual flat body, as well as two DevBot test cars. They, unlike the main car, have a cockpit and controls, however, the test machine has an identical set of sensors for sensing the environment, so developers can relatively easily transfer the skills of algorithms between two types of machines.
While the organizers conduct test races and competitions on test cars, as well as compare their work with the skills of professional pilots. However, in 2021, competitions on fully driverless cars should begin.