Inquiry About Multi-SLAM in Project Aria

[gogojjh]

Thank you for the incredible work on Project Aria! I’ve been exploring the platform and collecting data over a wide area, and I’m amazed by the quality of the multi-SLAM trajectories. It’s truly impressive.

I’m curious about the details of how multiple trajectories are aligned into a coordinate frame. I have several queries: Does the system use an approach similar to COLMAP, combining image retrieval and geometric verification? Does it leverage GPS information to narrow down the scope for image retrieval, making the process more efficient? Does it use sequential information (such as seqSLAM) to guarantee accuracy?

I would greatly appreciate it if you could share any insights or technical details about this.