Active Perception Workshop

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

We offer two ways to submit your method:

🔓 Public Submission

Submit your method via a pull request to the challenge repository.
Fill out the Google Form and include your pull request number.

🔒 Private Submission (code will not be shared!)

Send a .zip file of your full repository to activepws@gmail.com with the subject line [CHALLENGE]. Include a README describing your changes, especially if they go beyond the method folder.
Fill out the Google Form and mention the email address used to sent the .zip folder.

You can freely decide to submit privately or publicly, this does not affect the score. We believe that most of your changes should be contained in the method folder. We will run your models on the test sets and publish the results on the website. For public submissions, scores will also be posted as a comment to your pull request.