Motion as a Sensing Modality for Metric Scale in Monocular Visual-Inertial Odometry

Researchers Hadush Hailu and Bruk Gebregziabher have published a breakthrough paper demonstrating that specific robot motion patterns dramatically improve the accuracy of monocular visual-inertial odometry (VIO). Their key insight is that translational acceleration—generated by curved trajectories—is the fundamental source that couples metric scale to the inertial state, not the constant-speed straight-line travel commonly used. This relationship is formalized through gravity-acceleration asymmetry in IMU models, leading to new rank conditions on observability matrices and a lightweight excitation metric computable directly from raw IMU data.

Controlled experiments on a differential-drive robot equipped with a monocular camera and consumer-grade IMU validated the theory with striking results. Straight-line motion yielded 9.2% scale error, circular motion reduced this to 6.4%, and figure-eight motion achieved just 4.8% error—representing a 48% improvement over baseline straight-line performance. The excitation metric spanned four orders of magnitude across different motion patterns, providing a quantifiable measure of how effectively different trajectories resolve scale ambiguity.

This research establishes trajectory design as a practical, software-based mechanism for improving metric scale recovery without requiring additional hardware or sensor fusion. The findings have immediate implications for robotics, autonomous vehicles, and AR/VR systems that rely on accurate 3D positioning from limited sensor suites. By simply programming more dynamic motion patterns, engineers can extract significantly better performance from existing camera-IMU combinations.