olivaw-slam

Architecturally this does what slam_toolbox does inside ROS2 — but as a plain Rust library. It runs on macOS, Linux, and anything else Rust targets, and cross-compiles to a Raspberry Pi or Jetson with a single cargo build --target aarch64-unknown-linux-gnu. No middleware, no distro pinning, no C++ toolchain.

How it works

Feed it lidar scans — from olivaw-lidar or any other source — and it returns a consistent occupancy-grid map and a pose estimate.

Scans are preprocessed (gated, outlier-filtered, voxelised), then matched by a correlative matcher against the accumulated map rather than against the previous scan. That single decision is the core of the design: matching to the map means drift does not compound the way it does with frame-to-frame odometry.

Keyframes are taken every 0.3 m or 0.3 rad and feed three things at once: the log-odds occupancy grid, a pose graph built on factrs, and loop-closure detection. An accepted loop closure becomes a constraint in the pose graph, the graph corrects every pose, and the corrected poses redraw the grid — which is then what the next scan matches against.

Why it exists

ROS2 is a reasonable answer if you are already inside it. If you are not, the cost of adopting it to get one algorithm is enormous: a distribution to pin, a build system to adopt, and a C++ dependency tree to maintain. A library that is just a library has none of that.