// the take
maplab 2.0 is ETH Zurich's open research framework for visual-inertial SLAM, multi-session/multi-robot mapping, loop closure, and dense reconstruction, built as a set of modular C++/ROS packages. It's aimed at robotics researchers and grad students building or extending mapping pipelines, not at teams wanting a drop-in production SLAM stack.
It covers a genuinely wide pipeline in one place — VIO (ROVIOLI), multi-session map merging/optimization, vocabulary-based loop closure with its own hnswlib/inverted-multi-index implementations, and voxblox-based dense reconstruction, so you're not stitching five separate research repos together. The package split (ceres-error-terms, imu-integrator-rk4, feature-tracking, loop-closure-handler, etc.) means you can lift one module, like the IMU preintegration or the loop-closure engine, into your own stack without pulling in the whole console app. It's backed by two published, peer-reviewed papers and has actually run on multi-robot field deployments, which is a real signal beyond typical academic-code polish.
The documented install path is pinned to ROS melodic on Ubuntu 18.04/20.04 — porting to ROS2 or a current Ubuntu LTS is on you, and that's a nontrivial lift given how deep the ROS coupling goes. Last push was mid-2024 with no tagged release since the 2022 maplab 2.0 paper drop, so this reads as maintained-when-a-student-needs-it rather than actively developed; don't expect fast issue turnaround. The build pulls in a large tree of git submodules plus binary vocabulary/projection-matrix files checked straight into the repo, which makes for a slow, fragile first build and a bad fit for any CI pipeline. Tutorials and API docs live entirely on an external university-hosted wiki rather than in the repo, so there's real link-rot risk once ASL stops paying for that infrastructure.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →