On September 17th 2018, PIX launched the First Self-driving Engineer Training Base in partnership with Udacity China and Tier IV, one of the top self-driving companies in Japan who runs Autoware Academy. Under the joint efforts of the three founding members, the training base located in Guiyang will be dedicated to providing onsite training and test for self-driving engineers.
Why The Training Base
Self-driving engineers are badly needed in the self-driving industry. Almost every self-driving startup is chasing after the talented engineers. The cultivation and growing of engineers are highly dependent on the real vehicle test and algorithm optimization. However, the self-driving test sites around the globe are less than 20 and the actual onsite training base with mentorship is close to zero. Besides, the repeated onsite test and engineering are the essential steps for profound development of self-driving technologies.
Founded by PIX, Udacity and Tier IV, the Self-driving Engineer Training Base enables engineers to experience the fun of engineering and testing from self-driving mechanics, electronics to algorithms, helping improve their development abilities of both software and hardware. By constantly testing codes and algorithms on real car and real roads, engineers and startups will master the principles and mechanism of self-driving, bringing advancement and development of autonomous driving regarding its safety and use cases.
What Will Be Provided and Experienced
The Self-driving Engineer Training Base is equipped with self-driving hardware, facilities, devices and test sites with road deployment and more so that self-driving startups, engineers and students alike can do hands-on test & training for their self-driving technologies as well as algorithms.
- Factory With Cars and Tools
The 5000 ㎡ PIX factory is abundant with various tools, devices and facilities for self-driving car retrofitting, hands-on debugging and simulation test. There are four cars with drive-by-wire capabilities, which are available to engineers for onsite training and test. By 2019, the number of cars will increase to more than 10. Basically in PIX factory, self-driving engineers will be able to build a self-driving car from the scratch, enjoying the full-stack experience from hands-on work to software/algorithms optimization on real cars.
- Industrial Park With Test Deployment
Besides PIX factory, an industrial park in Guiyang with self-driving test course is also available, which is deployed with test facilities such as traffic lights, fake pedestrians, Bus Stop sign etc., and also multiple-laned city roads with controlled access. PIX test site in the industrial park provides everything needed for self-driving car testing on real roads.
The self-driving Training Base offers mentorship and guidance from self-driving software and hardware experts during the onsite courses to ensure the outcomes. Currently there are many enterprises in the self-driving industry who have become partners for the training base such as Udacity China, Autocore, Tier IV, Velodyne and Neousys, more is coming, which means additional hardware and software support.
Current Stage and Future Plan
Currently the Base is greeting the very first cohort of 15 engineers from Udacity China. Today is the third of the Five-day Onsite Self-driving Training Course, a trial course to kickstart the base launching and get more feedbacks for better user experience in future course. The Self-driving Engineer Training Base will be open to the public soon, providing hardware, site, mentorship and other support for global self-driving engineers, students and startups to test and optimize their self-driving technologies, further driving the development and safety within this industry.
Quick Catch-up on the trial training course which is On for the 3rd day
On Sep.17th, launch event of the Self-driving Engineer Training Base in Guiyang, the Data Valley of China. Then launch even was a great success with keynote speech, workshop, open ceremony, and networking buffet. Keynote speeches topics ranged from self-driving talents cultivation, LiDAR technology, open source software to the background story of the training base, presented brilliantly by Prof.Takeda from Tier IV, who is also professor and Lab Lead at Nagoya University, Xiong Shangwen, CEO of Udacity China, Mei Jiawei, senior engineer within Velodyne, and Yu Chuan, CEO and Founder of PIX self-driving project.
Engineers were excited about the coming onsite course because they could apply their codes and programs on the real car and make it move, which was a meaningful thing by itself. The 15 engineers were selected out of hundreds of applications by Udacity. They are from cities all around China: Shanghai, Beijing, Shenyang, Xi’an, Hubei and more. Even the hurricane in Guangzhou didn’t stop their enthusiasm from participating the training base and all engineers arrived on time before Sep.18th, the day when the trial training course officially started.
Self-driving Training Base Day #1
The non-stopping rain in Guiyang on the first day seemed to bummer the starting, but it was not. The whole morning was about the lecture on self-driving software from Prof.Alexander, Doctor of Science, Consultant at Tier IV and professor at Nagoya University. It was great honor to have Alex here as he is an expert in self-driving software Autoware and 3D mapping fields. Autoware is one of the most popular open source self-driving software and it’s widely used among engineers and developers in the self-driving industry. With his mentorship, engineers had a basic understanding to this system. In the Capstone project of Udacity Self-driving Engineer NanoDegree, Autoware has become the bridge from software to hardware for global students.
PIX has provided three by-wire control cars with Velodyne LiDAR VLP 16 for students to have hands-on test, and they needed to divide into three teams. Nothing got more interesting than grouping: writing down names, sharing skills, “selling” oneself out to the desired team members. It really took a while and finally we had the three teams, who could’t wait to jump on the cars just the minute after lunch. For self-driving software engineers, the excitement from touching and connecting with the physical cars and running the car on the test site with their own algorithms was the same feeling as when an talented musician finally got his beloved instrument and then played a splendid piece of music in front of the like-minded audience. Well, what kind of “music” will the three groups of engineers program and code on?
Self-driving Training Base Day #2
On the first day, the teams successfully finished the software and system deployment on the hardware of each car: TensorFlow, ROS, Autoware and so on. The deployment may sound boring and time-consuming, but actually most problems might happen and get stuck in the process of this step, just like how important engineering is while building a car. The process of installing and deploying software was the experience to explore and get more familiar with the various software tools.
On Sep.19th, Prof.Alex gave a profound introduction about how to use Autoware on self-driving cars, including mapping, data collection, labeling and detection etc..The framework of Autoware was a bit complex and it would take two mornings to explain the whole structure so that students would have a clear understanding. It was great and engineers responded that they had learned a lot. Even for engineers who had used Autoware before, they gained more tips and methods for some specific approaches while using this system.
In the afternoon, the three groups all went to one of the test sites in Qilin Plaza nearby, where they made data collection and traffic detection preparation, and then went back to PIX factory for model training and classifier test. All went great and the warm sunlight really facilitated things. Each group had a clear division and collaboration of tasks. The whole process was efficient though sometimes there was trouble, Prof.Alex and PIX staff were there to help. Night shadowed yet the passion of engineering continued.
Self-driving Training Base Day #3
Stayed up and rose up. The morning of Day #3 greeted the last part of the lecture. After working together for two days, engineers and mentors were getting along, a lot of questions and discussions during the lecture, which lasted till lunch time.
Each team had collected hundreds and thousands of pictures of the traffic lights from different position and angel. Big data from Day two’s four-hour onsite collecting. Manual labeling was a tedious but essential work. It was the best way to interpret the data. By carefully observing the images and data, each team could clearly see the quality of collected data.
The process of labeling gradually indicated the characteristics of each team. They tried various approaches, such as “production line” mode for label processing, and “balanced” mode fo labeling. Fine-tuning came after the labeling. Fine-tuning is the fastest way to achieve good test results currently. Migration learning can take advantage of the world’s best- detected and fastest network models. After certain training, the best model was exported and embedded in the node detected by Autoware. The next step would be the most thrilling: onsite test.
The five-day trial course of Self-driving Engineer Training Base has passed three days, so far so fantastic! In the next two days, each team would keep training the model, build maps, have onsite test for traffic lights recognition, add codes to Autoware and optimize. The final day there will be a fun “gaming” session to see which team brings better results. Surprises lie ahead, plus party and beers!
The Self-driving Engineer Training Base will be open to the public soon. Follow us and stay tuned for more updates and training course highlights sharing.
Medium: PIX Moving
Community Slack: https://movingcommunity.slack.com
Facebook Page: https://www.facebook.com/pixmoving