Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Learn about Rectified Point Flow, a unified approach to solving geometric alignment problems in robotics through this 50-minute conference talk. Discover how this innovative method casts point cloud registration and multi-part object assembly as a single conditional generative task by learning continuous point-wise velocity fields that transport unposed points to their target locations. Explore how this approach naturally recovers part poses and captures object symmetries without supervision, demonstrating superior performance across six benchmarks compared to traditional task-specific pipelines and explicit pose regression methods. Gain insights into practical applications in cultural heritage preservation, where robust alignment and assembly techniques are essential for documenting, restoring, and reassembling fragmented artifacts and monuments. The presentation covers the theoretical foundations of the method, its advantages over existing approaches, and real-world implementation scenarios that bridge civil engineering, architecture, and machine perception for creating data-driven sustainable environments.
