Ross Finman, a PhD student in Prof. John Leonard’s group at CSAIL at MIT working on object segmentation and 3D scene understanding will be visiting later this week and talking about his work over the past few years. Hope to see you there.

Tom

—

VIP Group Seminar
Object Discovery and Segmentation in Dense RGB-D Maps

Ross Finman
Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT

————————————————–
Date: October 9th, Thursday
Time: 14:00
Room: Huxley LT 144 (Lecture Theatre)
————————————————–

Robots have finally escaped from industrial workplaces and are making their way into our homes, offices, and public spaces. In order to realize the dream of robot assistants performing tasks together with humans, we need to provide them with the capability of understanding complex, unstructured, large-scale environments and the objects within them. Ideally this understanding is developed autonomously as not to burden a user. In this talk we present our work on enabling robot systems to discover, segment, and detect objects in the context of the environment the systems are working in, as well as using those object for improved mapping. Having robots learn objects from their environment instead of from large databases provides a more adaptive system to function in the complex real world. Specifically, we use motion cues in dense RGB-D maps to discover objects in their environment, thus learning the objects that are most frequently used. We then discuss how to use these discovered objects to optimize multiple scene segmentations in order to segment the object on future traverses. We will cover methods for segmenting maps in an efficient, real-time framework. Lastly, we will discuss current work on using place recognition for automatic object discovery.

Short Bio
Ross Finman is a Ph.D. Student in Electrical Engineering and Computer Science at the Massachusetts Institute of Technology. His research is focused on the intersection of computer vision and robotic mapping applications by using theory developed for the former in the new data domain of the latter.