| Title | : | Deliberative Perception for Multi-Object Recognition |
| Speaker | : | Venkataraman Narayanan (CMU, USA) |
| Details | : | Thu, 27 Oct, 2016 2:00 PM @ BSB 361 |
| Abstract: | : | A fundamental machine perception task is to identify and localize objects of interest in the physical world, be it objects on a warehouse shelf or cars on a road. Most modern-day methods for object instance localization employ scene-to-model feature matching or regression/classification by learners trained on synthetic or real scenes. While these methods are typically fast in producing a result, they are often brittle, sensitive to occlusions, and depend on the right choice of features and/or training data. In this talk, I will argue that deliberative reasoning--such as understanding inter-object occlusions--is essential to robust perception, and that the role of discriminative algorithms should mainly be to guide this process. Towards realizing deliberative perception, we have developed two algorithms: PErception via SeaRCH (PERCH) and Discriminatively-guided Deliberative Perception (D2P). PERCH formulates multi-object recognition as a search for the best global explanation of the observed scene, while D2P extends PERCH by allowing modern statistical learners such as deep neural networks to guide the global search. The latter is made possible by Multi-Heuristic A* (MHA*) and its extensions, graph search algorithms which we developed for handling multiple, possibly "inadmissible" heuristics. These algorithms allow us to leverage arbitrary learning-based algorithms as heuristics to accelerate search, without compromising on solution quality. Finally, I will present experimental results on real-world benchmarks, and from our participation in the 2016 Amazon Picking Challenge to emphasize the importance of deliberative reasoning for robust real-world perception. Speaker Bio: Venkatraman Narayanan is a Ph.D. student at the Robotics Institute, Carnegie Mellon University, where he is advised by Maxim Likhachev. His research interests include 3D computer perception, artificial intelligence, and robot motion planning. He has published in top robotics and AI conferences including RSS, ICRA, IROS and IJCAI, has won a best poster award at the 2014 International Symposium on Combinatorial Search, and is one of the recipients of the 2014 AAAI Robotics Fellowship. He has also spent time working on self-driving cars at the Uber Advanced Technologies Center and at Google X. Prior to Carnegie Mellon, he obtained his B.E. in Electronics and Communication Engineering from College of Engineering, Guindy in 2011. |
