Tightly-Coupled vs. Loosely-Coupled Fusion

Two architectures for multi-sensor fusion: combining raw measurements together, versus computing each sensor’s own result and merging afterward.

This distinction comes up constantly in multi-sensor fusion (such as camera+IMU or lidar+IMU). In loosely-coupled fusion, each sensor first computes its own result independently — for example, visual odometry computes a pose on its own, which is then weighted and merged with the pose from integrating the IMU inside a filter; this is simple to implement and its modules are swappable, but if one sensor fails on its own (say, tracking is lost from too little texture), its bad result feeds straight into the fusion. Tightly-coupled fusion instead puts raw measurements — such as the reprojection error of image feature points and IMU pre-integration terms — directly into one shared state estimator for joint optimization or filtering; this uses the information more fully and is usually more accurate and robust, at the cost of a more complex implementation and heavier computation. Systems such as VINS-Mono, OKVIS, FAST-LIO2, and LIO-SAM are all tightly-coupled designs.

ExampleA drone flying past a blank white wall has almost no visual features: a loosely-coupled system’s visual pose estimate can go badly wrong, while a tightly-coupled system can still maintain a good estimate using the few available feature points plus the IMU constraint.

Also called

Tightly-Coupled Fusion, Loosely-Coupled Fusion

Related

Multi-Sensor Fusion · Visual-Inertial Odometry · LiDAR-Inertial Odometry · IMU Preintegration · VINS-Mono / VINS-Fusion · Extended Kalman Filter

Sources

VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator (arXiv 1708.03852)
LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping (arXiv 2007.00258)