2nd CPM Olympics

Prize sponsored by INFORM

We would like to thank all 6 participants. All in all, it was an exciting competition with 28 submitted planners. We are looking forward to seeing you all in the next Olympics.

The 2nd CPM Olympics is divided into two stages, and only the second stage is relevant for the selection of winners.

The first stage serves as training and preparation. Participation in this stage is not mandatory but highly recommended. This year, in addition to the 16 standard training scenarios, we offer all the unhidden benchmark scenarios of the 1st CPM Olympics to help you develop your planners. They cover various traffic situations such as traffic jam, crossing pedestrians and bicycles on the road, to name a few.

In the second stage, we will publish 15-30 benchmark scenarios. They will only include the traffic situations contained in training scenarios. You have two weeks to make final adjustments to your planner and submit it for the official ranking. We allow multiple submissions and the best one will be considered while choosing the winners; despite this, please take each submission seriously, as the computational resource for evaluation is limited and participants with fewer number of submissions will have higher priorities. To avoid tailored solutions, we will hide 40-50% of the benchmark scenarios, meaning that your planner’s performance in these hidden benchmark scenarios will stay unknown until stage 2 ends.

Below is a summary of important information:

Stage 1 – Training and Preparation: now–Apr. 10, 2024

• Participation is not mandatory but highly recommended.
• 16 training scenarios plus 13 unhidden benchmark scenarios of the 1st CPM Olympics are available to develop your planner.
• This stage does not influence the selection of winners.

Stage 2 – New Scenarios: Apr. 10, 2024–May 8, 2024

• The release of 15-30 benchmark scenarios. The selection of winners is based on your planner’s performance in these scenarios.
• 40-50% of the benchmark scenarios will stay hidden to avoid tailored planners.
• The ranking status of public submissions and their corresponding scores in the unhidden benchmark scenarios will be periodically updated and made available on our leaderboard. Upon the completion of stage 2, comprehensive information pertaining to the selection of winners, including the scores of all submissions across all benchmark scenarios, will be published on the leaderboard to ensure transparency.

PS: All times are in Central European Time.

The CPM Olympics contains real-world scenarios that comprise intersections and roundabouts. These scenarios were extracted from the Roundabout Dataset ( rounD) and the Intersection Dataset ( inD), both of which possess restricted rights that prohibit direct downloading. Nevertheless, they can be simulated and benchmarked within our CPM Remote environment. Most scenarios are annotated with tags (such as Bicycle, Pedestrian and Bus) that describe the traffic situations and obstacles encountered during executions. Benchmark scenarios only constitute a subset of tags of the training scenarios, making them theoretically solvable by well-trained planners.

Your objective is to design a motion planner capable of generating feasible trajectories for navigating vehicles from given starting positions to appointed destinations. The primary aim is to minimize cost while satisfying all hard requirements (refer to the “Evaluation” section for further details). We restrict each scenario within a rectangular boundary indicated by yellow sides. Your planner governs the movements of all vehicles entering the scenario, while additional traffic participants, such as pedestrians and cyclists, act as dynamic obstacles. Real-time information on their current positions is available, but their intended trajectories remain unknown to you.

Submissions will be open once stage 1 starts. You are allowed to submit multiple times, while only the highest-performing submission will be considered for the leaderboard ranking. Additionally, you can choose to mark your submissions as either private or public. By selecting the private option, the submitted results will not be publicly ranked, and only you will have access to them. On the other hand, if the public option is chosen, your submissions will be considered by the leaderboard after undergoing the benchmarking process.

The evaluation of a submitted planner in each scenario involves two distinct parts. The first part verifies the fulfillment of all specified hard requirements, and the failure of this part will result in a score of zero for the given scenario. The second part establishes metrics to assess the quality of the planner, resulting in a weighted score. The overall score is determined by averaging the scores obtained across all scenarios.

You can either participate as group (with unlimited number of members) or as solo participant. Please note that team members are not allowed to participate in multiple teams and/or submit as solo participant. As a thorough check will take place before the winners’ announcement, make sure to use your real name when registering. Reward may be rejected if you cannot prove your identity.

The winners will get:

• The opportunity to present their algorithms at our  workshop during the IEEE Intelligent Vehicles Symposium 2024 in South Korea, through either a 4-minute oral presentation, a 2-minute video presentation, or a poster accompanied by a 2-minute video presentation.
• Prize money from a total pool of €6,000 (approx. $6,000)
• Certificates recognizing their achievements.

To be eligible for these awards, the winners must:

• Agree to have their names announced on our website.
• Send us the 4-minute video presentation or the poster accompanied by a 2-minute video presentation before May 20, 2024 if they are unable to attend our workshop for an oral presentation.

• All rights of the submitted code will stay with the author. We neither publish the code nor use it ourselves.
• The cost function does neither reflect reality nor claims to be complete. Planners should be able to deal with different demands.