In cuVSLAM, visual and inertial measurement unit (IMU) residuals are combined during sparse bundle adjustment. The balance between the two sensor streams is set by fixed weights derived from pre-calibrated noise covariances. This leaves no mechanism to adjust the visual-inertial balance during operation.

The paper, titled "Smart Fusion for Adaptive Visual-IMU Weighting in cuVSLAM Sparse Bundle Adjustment," addresses this limitation by proposing an adaptive weighting mechanism. The title and abstract indicate the goal is to make the fusion "smart" so that the weighting can change rather than remain static.

The available abstract is truncated, so the specific method for computing adaptive weights is not described in the source. The significance is clear: if successful, the approach would allow cuVSLAM to respond to changing conditions, such as visual degradation or IMU drift, without requiring manual re-calibration.