Challenge

Participants are provided with a sparse map of an environment and a simulated robot equipped with a camera that can rotate independently of the mobile base. Given a set of robot waypoints, the objective is to orient the camera toward the most informative parts of the environment to improve localization accuracy.

Full challenge details are available at: github.com/rvp-group/actloc_benchmark. If you have any questions or encounter issues, please open an issue on the repository—this helps others as well.

There are two challenge tracks:

• Single-view active localization – predict the best camera orientation at given robot positions.
• Path-level active localization – jointly plan camera and robot trajectories for improved localization.

Additionally, a performance-based award will be given to the fastest high-performing method.

To ensure a fair comparison across all participants, we will evaluate your submissions on a commercial workstation with the following specifications: RTX 4090 (24 GB VRAM), 64 GB RAM. Please ensure that your method fits within these hardware constraints. Submissions that exceed these memory limits may be disqualified.

Submission

Submission Deadline: 21st September 2025, 23:59 AoE

• Submit your method via a pull request to the challenge repository.
• Fill out the Google Form and include your pull request number.
• Most of your contributions should go into the method folder. We will run submitted models on the test sets and publish results on the website. A comment with your scores will be posted to your pull request.

Leaderboard

The leaderboard will be published here as submissions are evaluated.