Warehouse aisles, uniform walls, and specular floors are notoriously difficult for point-feature visual odometry, yet these same environments are rich in straight edges: rack uprights, door frames, and wall junctions. The authors of a new arXiv preprint extend cuVSLAM, a CUDA-accelerated visual-SLAM system, to exploit those linear structures directly. Lines are represented as Plücker coordinates for geometric transport and as a minimal 4-DOF orthonormal form for optimisation, with a Cauchy-robust point-to-line residual folded into the per-frame Gauss-Newton pose solve and the sliding-window bundle adjustment.

A key design choice is that the line extension adds residual blocks to existing solvers rather than introducing a second estimator. An adaptive information weight lets lines assume more of the estimation burden exactly when point features become scarce, and disabling the extension reduces the system to the unmodified point pipeline. This means the added robustness comes without a separate tracking mode or a fundamental change to the underlying architecture.

Evaluated in RGB-D mode with SE(3) Umeyama alignment and no scale correction, line augmentation reduced mean absolute trajectory error by 38.5% over the 22 sequences of the TUM RGB-D Freiburg3 collection and by 30.1% over the 8 Replica room and office sequences. Median translation error fell by 38.9% and median rotation error by 66.4%, with zero tracking failures in any configuration tested. The results suggest that straight-edge structure, often dismissed as ambiguous, is a reliable signal for visual odometry in precisely the settings where point features fail. This is a single-source report; no independent verification or contrasting results are included here.