Event Location: NSH 1507

Abstract: In many real-world applications, motion estimation and mapping must be conducted online and in real-time. The real-time motion estimates are important for controlling and maneuvering autonomous vehicles, and the maps generated online are critical for obstacle avoidance and path planning. Further, the final map can be used as a representation of the traversed environment or taken as input for further processing such as for scene segmentation and 3D reasoning.

This thesis proposes to combine range, vision, and inertial sensing to fulfill the aforementioned tasks. The proposed sequential processing pipeline includes three components: first, IMU mechanism is used for motion prediction; second, a vision based method runs at a high frequency to handle rapid motion and register point clouds; finally, a scan matching based method further refines motion estimates and point cloud registration to build maps. The three components repetitively solve and refine motion from coarse to fine. The system therefore is able to achieve accuracy in the level of off-line, batch methods while running online and in real-time.

In the proposed work, I will study robustness of the motion estimation in large-scale and extreme conditions such as in low-light, texture-less, or structure-less environments. I will also improve the system to handle high-rate rotation and translation. The final system will deliver precision and low-drift in motion estimation and high reliability w.r.t. environmental degeneracy and aggressive motion.

Committee:Sanjiv Singh, Chair

Martial Hebert

Michael Kaess

Larry Matthies, Jet Propulsion Laboratory