ADL Track
Autonomous Driving Lab has a cooperation project with Bolt to evaluate the state of self-driving technology. Bolt has sponsored us to buy a Lexus RX450h vehicle fully equipped with sensors and drive-by-wire technology.
As part of that project we are trying to make our test car to drive in real city traffic. We chose Autoware.AI open-source software as a basis for that experiment, for its use of already established technologies (ROS) and large user community.
In practice we have experienced many difficulties and bugs with Autoware, especially with its planner component called OpenPlanner. Therefore we would like to evaluate other open-source options for self-driving.
In this course you will have hands-on experience with one open-source self-driving software. As part of that, you will also learn some theoretical background in autonomous vehicle design and operation. We have chosen following softwares for the evaluation:
- Autoware.Auto (new version of Autoware)
- Baidu Apollo
- NVidia DriveWorks
- Comma.ai Openpilot
Organization
The course is organized such that we will divide students into teams and each team will take one software (if we have too few students, we can skip some of the softwares). Each of the softwares has accompanied online lectures that each team should watch independently. Every week we have a class meeting dedicated on one topic. Everybody should watch the lecture on the topic at home or preferably together with the team. One student prepares a test on the topic for others. In class we all take the test and after the test there is discussion of right answers.
The online materials to learn from are as follows:
- Autoware.Auto: https://www.autoware.org/awf-course
- Baidu Apollo: https://apollo.auto/devcenter/devcenter.html
- NVidia DriveWorks: https://developer.nvidia.com/drive/training
- Comma.ai Openpilot: https://medium.com/@comma_ai,
https://medium.com/@chengyao.shen/decoding-comma-ai-openpilot-the-driving-model-a1ad3b4a3612,
https://arxiv.org/abs/2003.06404
Each team is assigned a supervisor from Autonomous Driving Lab.
Teams
| Autoware.Auto | Baidu Apollo | |
|---|---|---|
| Team | Dmytro Zabolotnii, Kent Zuntov, Ivan Hladkyi | Romet Aidla, Enlik, Tetiana Shtym, Navid Bamdad Roshan |
| Supervisor | Maxandre Ogeret | Jan Aare van Gent |
Tests
Tests can be created using Google Forms. Tests need to be sent to supervisors by Monday morning on the week the test is performed, for quality assurance.
Schedule
All the class meetings are in Delta room 1022 on Wednesdays 14:15-15:45.
| Date | Topic | Student | Autoware lectures | Apollo lectures | NVidia webinars |
|---|---|---|---|---|---|
| 9.09.2020 | Course overview (slides) | Tambet | |||
| 7.10.2020 | Intro to self-driving | Tambet | Lecture 5, Lecture 6 | Lecture 1 | Clip, Webinar 1, Webinar 2 |
| 14.10.2020 | Mapping | Lecture 14 | Lecture 2 | Webinar | |
| 21.10.2020 | Localization | Lecture 10 | Lecture 3 | Clip | |
| 28.10.2020 | Perception: LIDAR | Lecture 7 | Lecture 4 | Clip, Webinar 1, Webinar 2 | |
| 4.11.2020 | Perception: Camera | Lecture 8 | Lecture 4 | DRIVE Networks, Webinar 1, Webinar 2 | |
| 11.11.2020 | Perception: Radar | Maxandre | Lecture 9 | Clip | |
| 18.11.2020 | Prediction | Lecture 5 | Webinar, Clip 1, Clip 2 | ||
| 25.11.2020 | Simulations | Lecture 11 | CARLA Talks 2020 | DRIVE Constellation | |
| 2.12.2020 | Planning | Lecture 12 | Lecture 6 | Webinar | |
| 9.12.2020 | Control | Lecture 12 | Lecture 7 | Webinar, Clip | |
| 16.12.2020 | Hardware | Lecture 4 | Webinar 1, Webinar 2 | ||
| 23.12.2020 | Infrastructure | Lecture 13 | Webinar, Clip |
Choose the topic you would like to create a test for here.
Deliverables
To earn 6 ECTS from this course each participant should:
- Create a test on one topic for other students and grade it.
- Get at least 60% of points from all tests created by other students.
- Demonstrate the respective software running on their computer against a simulator.
- Each team has to demonstrate the software running on Lexus car by driving a short pre-defined trajectory on parking lot. In case that turns out to be too complicated, the team should produce written report of their experiments and failures.
Follow-up
There is also a follow-up course planned in the next semester, where we make the teams compete to finish a demo track on the parking lot with the least amount of safety driver interventions.
Contact
Autonomous Driving Lab (room 3095)
Tambet Matiisen
tambet.matiisen@ut.ee