(c++) Developed a 3D HD map creation algorithm based on graph-based SLAM algorithm (Simultaneous Localization and Mapping with graph optimization).

• Designed a submap forming method to simplify the calculation unit from point-cloud to submap. Designed an adaptive trigger based on LiDar covariance for the execution of submap forming. Increased the map creation speed by 20%.

• Implemented Boost RTree algorithm to find loop-closure. Improved the RMSE by 18% compared to KdTree algorithm.

• Designed an outlier remover to check the global consistency of every loop-closure finding based on chi-square test.

• Utilized Google Protocol Buffer to serialize point-cloud pose related data. Speed up the data processing.

(MATLAB) Presented a VO method (Visual Odometry, algorithms estimating the robot state with vision information) based on direct intensity from RGB-D image.

• Implemented Gauss-Newton algorithm to solve the optimal transformation between frames by minimizing the intensity residual from all pixels. Increased the reliability of registration when camera distort/defocus.

• Adopted a sliding window optimization to process frames in batch. Realized a semi-global pose synchronization.

• Developed a coarse-to-fine data scheme to reduce the computation load in large camera motions (x2 speed up).