online finals

The "2020nd Japan Automotive AI Challenge Final" will be held from September 9rd to November 23th, 11.
The competition was scheduled to be held on June 6th this year at the Chiba Experiment Station attached to the Institute of Industrial Science, the University of Tokyo (on the Kashiwa Campus), but was canceled due to the spread of the new coronavirus infection.
Participants will submit programs and compete based on the results of the simulation.

Teams participating in the 2nd Japan Automotive AI Challenge online final competition (in order of entry)

Shallow Learners
Self-driving car study session
MTL*
tomo123*
CHD
NWC-WIN

MPRG
🔧🌟@
cats
Windridge
Yurara*
DeepLearners

*Participating from the 2nd AI Edge Contest. Other than that, the top teams in the 2nd Japan Automotive AI Challenge online simulation qualifiers

Competition outline

With the demand for food delivery services rapidly increasing due to the coronavirus pandemic, delivery by self-driving vehicles is expected to become a reality from the standpoint of logistics and disaster prevention. In this competition, participants use an online simulator to compete using self-driving cars to complete several tasks and deliver ordered items quickly and carefully without damaging them.

Announcement of results

The results will be announced in mid-December, and an awards ceremony (partially online) will be held.
Information on interviews will be sent to you at a later date.
In late December, we will create a running video from the programs created by each team and release a video of the cars virtually running side by side with commentary. Interviews with participants will also be made available.

Rules regarding the entire competition

You will be required to create source code that can run through scenarios using the open source simulator (LGSVL) available in Autoware.
The scenario is a time attack competition in which participants compete for the time from the start point to the goal point.
Participants create source code to clear a given scenario and first verify it in their local environment.
By uploading the source code to an online environment, the simulation will be performed online and the time will be determined. Ranking will be determined based on the simulation result time of the last uploaded source code. (For example, if the first upload result is 1 minutes, the second is 5 minutes, and the third is 2 minutes, the last 3 minutes submitted will be used (not the second 3 minutes of the best score)

Explanation about the scenario

Drive through checkpoints on a predetermined route. Reverse driving or shortcuts are not allowed.
Vehicle speed is set to a maximum of 30km/h. If you exceed 30km/h, you will be disqualified.
The time limit is 5 minutes. If you exceed 5 minutes, you will be disqualified.
In addition to buildings that originally exist on the map, pedestrians and other vehicles appear as obstacles, and if you collide with an obstacle, you will be disqualified.
You can either stop temporarily to avoid obstacles, or you can avoid them while driving (without stopping).
Local simulation and online simulation differ in settings such as the timing, shape, location, and movement speed of pedestrians and other vehicles.

Driving route map

① to ⑥ are the main points in this scenario.
The route and each point are approximate and not exact.

The green sphere is the approximate location of the checkpoint.
You must pass these checkpoints in the order of your route.

Scenario introduction ① Avoiding pedestrians crossing the road and vehicles parked on the road

Pedestrians cross the road from right to left.
Additionally, there are vehicles parked on the road (not shown in the reference image on the left).

The local environment and online environment scenarios differ in the number of people, the color of their clothes, their speed, and the timing and location of their appearance.

Scenario introduction ② Turn left at an intersection to avoid vehicles coming from the left side

A vehicle is approaching from the left at the intersection.
There are traffic lights, but they are always blue.
Vehicles coming from the left will continue without stopping (although there is a stop line in front of the intersection).

Local and online scenarios differ in vehicle type, color, speed, spawn timing, etc.

Scenario introduction ③Turn left into the parking lot avoiding oncoming cars

An oncoming car is coming.

Local and online scenarios differ in vehicle type, color, speed, spawn timing, etc.

Scenario introduction ④ Driving in the parking lot avoiding parked vehicles

There are vehicles parked in the parking lot.

Local and online scenarios differ in vehicle type, color, number, parking location, etc.

Scenario introduction ⑤ Driving compatible with vehicles traveling at low speeds

A vehicle is traveling at low speed in the traffic lane. (You may be driving in the opposite lane (vehicles are patrolling).)
It is possible to avoid the vehicle, overtake it, or follow it.

Local and online scenarios differ in vehicle type, color, appearance timing, etc.

Scenario introduction ⑥Go straight through the intersection avoiding vehicles coming from the left and right

Vehicles are traveling in both left and right directions at an intersection. (The image on the left only shows vehicles from the right direction, but there are also vehicles traveling from the left direction.)
The traffic light is always blue.

Local and online scenarios differ in vehicle type, color, number, appearance timing, etc.

(Reference) Traveling of own vehicle

It is assumed that the vehicle is basically driven in the driving lane, but it is also possible to drive in the oncoming lane as long as it is within the roadway.
It is prohibited to drive anywhere other than the roadway or parking lot.

(Reference) Oncoming vehicle driving

There are some parts where oncoming cars are driving, but they never cross the center line.
Please note that there are no other vehicles in the parking lot.

Simulation in online environment

Instructions on how to upload will be posted in the README of the github repository for this tournament.
Due to the specifications of the simulator, even if you upload the same source code, you may not get the same results.
The number of online simulations will be limited to three times per day for each team. The count will be reset at 0:00 every day.
If a problem occurs in the online environment, we will promptly notify you on github (if the problem occurs between 6 PM and 9 AM, the next business day). Responses will be considered separately depending on the impact.

Questions and information update notices

If you have any questions, please ask via github issues. Generally, we will respond within two business days, but please be aware that if there are a large number of questions or if the questions are difficult to answer, we may be delayed or unable to respond.
If there are any updates to the information, we will update the README on github from time to time. We will create an issue to notify you of updated information, so please subscribe. Subscribe to be notified when updates occur (please turn on notifications).

Display results after completion

Rankings will be determined based on each team's final simulation time in an online environment. Please note that the results will be verified after the main race period ends, so the ranking will be determined once the verification is complete.
After the main tournament period ends, the secretariat will run a simulator on the codes of the top teams and compile the simulation footage as reference footage for the main tournament. Please note that due to the specifications of the simulator, the running time may not completely match between the online environment and the video creation time. (Rankings will be determined based on performance in the online environment).
*If you have any doubts, such as the results being significantly different between the local and online environments, please contact the secretariat before the end of the main competition period.

Other

Please note that this main competition is strictly a simulation competition, so the driving will not necessarily match the actual road environment and traffic rules.
Due to the simulation specifications, the same code will not give the same time every time.
Although we conduct verification in advance and work to reduce technical issues, please understand that issues that the secretariat has not anticipated may arise.

Recommended Environment

◎Autoware recommended environment
*“Autoware” is a registered trademark of The Autoware Foundation.

要素	Recommended specifications
OS	Linux(Ubuntu18.04)
CPU	Intel Corei7 (8 cores)
memory	16GB or more
storage	SSD 30GB or more
GPU	NVIDIA Geforce GTX 1080 or higher (if using nodes with GPU implementation)

Autoware setup (v1.13.0)
https://gitlab.com/autowarefoundation/autoware.ai/autoware/-/wikis/Source-Build

◎LGSVL recommended environment

Equivalent to the Autoware recommended environment above.
When starting at the same time as Autoware, 32GB or more memory and RTX2080 or higher GPU are recommended.
Since this simulator also runs on Windows, a separate environment may be prepared.

Development code rights and disclaimer
The rights to the code uploaded to the competition system belong to the code developer. JSAE and its subcontractors will use the electronically submitted data (code) only for the purpose of scoring competition tasks and verifying the results, and the developer will grant permission for such use. In addition, it shall be destroyed once the above purpose has been completed. Developers agree in advance that the code they submitted will be deleted by JSAE and its subcontractors after the qualifying round, and shall not raise any objections to this. JSAE and its subcontractors (including subcontractors for cloud environments, etc.) will handle electromagnetically submitted data (code) with the due care of a prudent administrator, but they will not be responsible for any damage caused by theft, hacking, or other force majeure. (Excluding cases caused by intentional or negligent behavior of the Society or its outsourced parties).