What is JetRacer?

JetRacer is an AI race car powered by NVIDIA Jetson Nano.

Since its introduction as the latest compact AI computer, NVIDIA Jetson Nano has made it possible to develop AI-enabled devices and embedded systems at unprecedented speed and low cost. The Jetson Nano developer kit is also widely used as an AI learning material, making it possible to easily touch machine learning and deep learning, and JetBot appeared as an AI car capable of full-fledged autonomous driving. From there, it evolved further as a race car, and the open source assembly method was released as "JetRacer"is.

It is the first time in the world to take up JetRacer in hands-on for companies!

The "JetRacer Hands-on Event" held three times in December 2019 was conceived and planned by NTT Com Engineering as an in-house training from around June 2019, and was held. Although the number of participants was limited to 100, the number of applications exceeded expectations, and the venue was filled with enthusiasm on the day of the event.

Hands-on began with a proposal from the planner, Mr. Atsushi Sasaki, Technical Director, Cloud Application Division, Cloud Application & Voice Department, NTT Com Engineering Co., Ltd.

Mr. Sasaki, who has been involved in the development of supercomputers in the past, said that when learning deep learning, it was difficult to get an image of it from books. The thought that "seeing is believing" was encouraged, and after seeing the JetBot seminar that our company cooperated in holding, we proposed to hold this event within the company.

Engineers working at NTT Com Engineering are engaged in network and cloud services and are exposed to network equipment and servers on a daily basis, but AI engineers are still few. It was thought that I would like you to learn deep learning on this occasion and be able to use AI as a tool. In this hands-on session, which was prepared with the help of people from the company's human resources department, he spoke about his passion for employee training and his desire to contribute to the social implementation of AI.

JetRacer kit used in hands-on

We asked Mr. Yo Sasaki, Board Director of GClue Co., Ltd., to be the lecturer this time, and he prepared a JetRacer kit for this hands-on.

[JetRacer kit configuration]

・Jet Racer kit

JetRacer's base is Tamiya's 1/10 scale radio-controlled car, with a deck equipped with Jetson Nano, WiFi router, camera, and mobile battery.

・Jetson Nano Developer Kit

・Camera unit

・Wi-Fi router

・Large-capacity, high-output mobile battery

There was a JetRacer kit for each person, and it was like a Christmas present.

hands-on flow

1. explain the overall flow
2. Assembling the JetRacer Kit
3. Connect to Jetson Nano with a browser via the WiFi router installed in JetRacer
4. Creating a dataset
5. Training Create an inference engine by training with PyTouch on Jetson Nano using the created dataset
6. Autonomous driving Test autonomous driving on the course with the created inference engine

First, let's assemble the JetRacer

In the morning session, we even assembled the JetRacer. Mr. Sasaki from GClue explained the background to the creation of JetRacer and the latest status of autonomous driving in an easy-to-understand manner for first-timers.

First, attach the Jetson Nano, camera, and WiFi router to the base radio-controlled car made by Tamiya with screws, connect the camera to the Jetson Nano, and then connect the control module of the radio-controlled car to the Jetson Nano. This completes the assembly of JetRacer. The moment I turned on the JetRacer on the desk and moved the accelerator and steering wheel with the controller, everyone was full of smiles and seemed to be enjoying the hands-on experience. Mr. Atsushi Sasaki's smile was very nice, saying, "Everyone has a nice face!"

Let's run JetRacer autonomously on a special course

In the afternoon session, we will create a training dataset and drive autonomously with an inference engine trained using the learning framework (PyTouch).

The dataset creation and training framework are done on Jetson Nano running Ubuntu. Specifically, it is done by operating a program that runs on Jupyter Notebook on Jetson Nano with a browser on a laptop computer.

Create a data set by shooting the course on a special circuit with the camera installed in JetRacer and instructing the direction to go to the shot image. Some people shoot a lot on curves and a lot on straight lines, so we asked them to create a data set by taking about 100 photos while doing trial and error, such as places off the course and in the middle of the course.

Using the created dataset, learn using the learning framework (PyTouch) on Jetson Nano to create an inference engine for autonomous driving. The inference engine derives the angle of the steering wheel relative to the camera image. It is a mechanism that JetRacer autonomously runs by the information.

There are JetRacers that run autonomously along the course, but many JetRacers seemed to go off the course, but even if things didn't go well, everyone was working on it with a smile.

Inference tends to work on the outer white line rather than the central yellow line, and it tends to be guided. was repeated.

Finals and future

In this day's training, three people were able to make a lap around the course. Some of the other riders went as far as they could, but many of them were unable to turn the hairpin corners well and strayed from the course. However, it was a very enjoyable training that everyone in the venue was happy to see how well they went around.

This December, a total of 100 people took part in three hands-on sessions. On January 16, 2020, the JetRacer final competition will be held by participants. The top few who lapped the same course faster in the time attack will be able to participate in the GTC JetRacer competition held in San Jose, USA in March 2020 on behalf of NTT Com Engineering.

This is not just educational content for employee training, but a dream project where employees who demonstrate their abilities can interact with engineers from around the world. Macnica would like to continue to help make this hands-on experience an opportunity for everyone to make new discoveries.

Summary

The JetRacer kit is specially prepared for hands-on use this time. If you want to assemble it yourself, please purchase the Jetson Nano development kit, and for other parts, you can refer to the instructions published on NVIDIA GitHub. If you would like to purchase the Jetson Nano development kit now, please do so here.

Click here to purchase the Jetson Developer Kit

*The developer kit sales page will be moved to the Switch Science website.

This hands-on event was held as an in-house training for NTT Com Engineering. If you are interested in conducting JetRacer hands-on in the same way, please contact us.