Low-Power Localization via Graph Optimization on Sparse Time-of-Flight Scans

Pose graph optimization plays a vital role in SLAM systems by turning local motion into a globally consistent path. However, existing approaches often rely on dense feature extraction, computationally expensive matching, or manually tuned loop closure thresholds, limiting their applicability to resource-constrained platforms and sparse sensing modalities. This paper presents an adaptive pose graph optimization method tailored for lightweight 8$\times$8 Time-of-Flight (ToF) scans. The proposed framework dynamically computes loop closure thresholds based on the statistical distribution of scan similarities, eliminating the need for manual tuning. The resulting pose graph is optimized using standard sparse Gauss-Newton methods with low computational overhead. Extensive experiments demonstrate that the proposed method achieves competitive absolute trajectory error (ATE) while significantly reducing computational cost compared to traditional feature-based or dense point cloud SLAM frameworks. Our approach offers a computationally efficient solution for real-time SLAM on embedded and low-power robotic platforms operating with sparse depth measurements.