IEEE Circuits and Systems Society Montreal Chapter, IEEE Solid State Circuits Society Montreal Chapter, IEEE Montreal Section, ReSMiQ – Regroupement Stratégique en Microsystèmes du Québec invite you to attend the talk by DR. YONG ZHAO. “With the rapid development of large language models (LLM) featuring low cost and high performance, they will soon be applied to robots, cars, drones etc. The so-called embodied intelligence technologies will be the next big market opportunity. However, when these robots need to conduct certain tasks, they require advanced 3D perception ability of the environment. Although various 3D perception technologies, such as LiDAR, are also developing rapidly, they, similar to some other active perception devices, have many problems and limitations. For example, it may not produce effective reflections on mirror surfaces, walls parallel to the line of sight, etc. Moreover, it only has a sparse point cloud. Besides, it lacks texture and color. Therefore, for robots serving thousands of households, binocular vision technology, similar to human vision, is one of the key technologies for robot perception. Binocular vision algorithms have experienced a leapfrog development from traditional stereo matching algorithms based on image processing to a new generation of algorithms based on learning with deep nets. Although very high performance has been achieved, the computational load is still relatively high for embedded devices. Therefore, further simplification and optimization of the stereo matching algorithms is needed to enable widespread application on low-cost mobile devices such as robots.” Speaker(s): Dr. Yong ZHAO Room: 2.184, Bldg: Concordia University – EV Building, 1515 St. Catherine, West, Montreal, Quebec, Canada