AIC Qualification Phase Results & Advancing Teams

Hello everyone,

It has been an incredible three months of the qualification phase for the AI for Industry Challenge (AIC), and the response and participation from the community have exceeded all our expectations. We are truly humbled by the prowess of the solutions developed and the collaborative spirit shared across the community. This challenge has been far more dynamic and exciting than we could have ever imagined!

It is my absolute pleasure to announce the top teams advancing to Phase 1.

Advancing Teams


Why 31 Teams Are Advancing

You might notice that there are 31 teams listed above, even though our initial guidelines stated that only the top 30 would advance.

We want to fully acknowledge a discrepancy in the closing times of our qualification submission portal. The submission portal form closed at 16:59 PM PT on May 15th, while our official Terms and Conditions (T&C) specified the deadline as 23:59 PM PT on that same day.

As a reminder, our submission workflow consisted of two steps:

  1. Uploading your OCI image to our registry.
  2. Submitting the unique URI of the OCI image into our submission portal.

While the registry continued to accept image uploads until 23:59 PM PT, the early closing of the submission portal prevented several teams from completing Step 2.

To ensure fairness and strictly honor our T&Cs, our evaluation team manually ran all images uploaded to the registry prior to 23:59 PM PT and updated the leaderboard accordingly. One of these late-uploaded images scored high enough to enter the Top 30. Rather than displacing an existing team that had already secured a spot before 16:59 PM PT, we decided to expand the roster to 31 advancing teams. As we say, more players = more fun!

For the teams that uploaded late, we will be reaching out to you individually to clarify your final scores.


Celebrating the Ecosystem

The true winner of this challenge is the robotics ecosystem. Every single one of you has helped lift the entire community. I am personally proud to see so many cutting-edge solutions built upon ROS 2 and its surrounding tooling to tackle this extremely difficult contact-rich cable insertion task.

To help teams analyze their results, we have released the exact aic_engine configuration used for the qualification phase evaluation. You can check it out in this Pull Request. Feel free to download this configuration and rerun your solutions locally to see how they perform.

Share Your Approach!

Lastly, we encourage everyone—whether you advanced or not—to share how you tackled the problem in this thread! Let’s learn from each other and continue to democratize access to advanced robotics together.

Good luck to the 31 advancing teams as they enter Phase 1!

7 Likes

I continued this work even after the qualification stage just for my personal learning. Here’s my Twitter thread

2 Likes

Thanks for the challenge @Yadunund!
We from Comsysto Reply participated as the team MacCody and were on a special exploration path: can a coding Agent with supervision by domain non-experts create meaningful results?
If you want to know more about our experiences and takeaways, have a look at this post. We are open to discuss further questions and ideas with the community!

We also like the idea that LLMs and Coding Agents can change the game for classical Robot Control Code instead of a policy that may need to be trained on hefty GPUs (although it’s important to say that the frontier LLMs are also a product of GPU Cluster training cycles).

We learned a lot and wish all further participants who got qualified great success in the following phases!

1 Like

the orginal schedule is

However,we have received the email

Is the phase 1 competition delayed by two weeks?

Will the phase 2 be also delayed?