Ashutosh Saxena, Assistant Professor, Microsoft Faculty Fellow, and Alfred P. Sloan Research Fellow in the Department of Computer Science at Cornell University, presents How Should a Robot Perceive the World?

Abstract

In order for a robot to perform tasks in the human environments, it first needs to figure out "what" to perceive. While for some robotic tasks (such as an object finding robot) this is relatively straightforward (e.g., infer the object labels from RGB-D data), many other robotic tasks require a robot to be more creative about what to perceive. For example, for a robot to arrange a disorganized room, it would need to perceive the human preferences about the usage of objects as well as the low-level manipulation strategies. In this talk, I will illustrate the issues surrounding "what to perceive" through a few examples.

The key to figuring out "how" to perceive lies in being able to model the underlying "structure" in the problem. I propose that for reasoning about the human environments, it is the humans that are the true underlying structure in the problem. This is not only true for tasks that involve humans explicitly (such as human activity detection), but also true for tasks in which a human was never observed! In this talk, I will present learning algorithms that model such underlying structure in the problem.

Finally, using such learned structure, I will present several robotic applications ranging from single-image based aerial vehicle navigation to personal robots performing tasks of unloading items from a dishwasher, loading a fridge, arranging a disorganized room, and performing assistive tasks in response to human activities.

Bio

Ashutosh Saxena is an assistant professor in computer science department at Cornell University. His research interests include machine learning and robotics perception, especially in the domain of personal robotics. He received his MS in 2006 and PhD in 2009 from Stanford University, and his BTech in 2004 from Indian Institute of Technology (IIT), Kanpur. He was a recipient of the National Talent Scholar award in India and the Google Faculty award in 2011. He was also named a Alfred P. Sloan Research Fellow in 2011, and a Microsoft Faculty Fellow in 2012.

Saxena has developed Make3D (http://make3d.cs.cornell.edu), an algorithm that converts a single photograph into a 3D model. Tens of thousands of users used this technology to convert their pictures to 3D. He has also developed algorithms that enable robots (such as STAIR, POLAR (http://pr.cs.cornell.edu) to perform household chores such as unload items from a dishwasher, place items in a fridge, etc. His work has received substantial amount of attention in popular press, including the front-page of The New York Times, BBC, ABC, New Scientist, Discovery Science, and Wired Magazine. He has won best paper awards in 3DRR and IEEE ACE, and he was named a co-chair of the IEEE technical committee on robot learning.

Abstract

During the next 10, 15, 20 years, society is going to look toward robotics for solutions to some of its big challenges: From improving the self-sufficiency and dignity of aging, to taking care of the next oil spill, passing through enabling cost-effective autonomous manufacturing, or actively supporting space and planetary exploration.

The ability to perform purposeful and reliable physical interaction is central to all these challenges. Unfortunately, our models of the mechanics of manipulation are usually based on simplifications and assumptions, which often fall short when applied to real problems and real scenarios. Enabling autonomous robotic manipulation in the real world requires to close the gap between models and reality, either by constructing more realistic models, or by quantifying the risk of failure of the inaccurate ones.

In the first part of the talk I will show recent work in the Manipulation Lab on constructing data-driven probabilistic models for manipulation and haptic perception that enable deriving manipulation plans with accurate predictions of the probability of success. These models are constructed with extensive offline experiments. I will illustrate the process with the example tasks of placing an object, dropping it in a cavity, or inserting it in a hole.

In the second part of the presentation I will talk about the most relevant aspects of the simple gripper used in the experiments, with an informed design that allows us to explore manipulation principles. In particular I will discuss our exploration of the problem of designing the shape of the fingers to satisfy some mechanical functionality. I will illustrate the technique with the example of designing the finger to improve the stability of a grasp.

Mark Dredze, Assistant Research Professor in the Department of Computer Science at Johns Hopkins University, presents Models for Mining Public Health Information from Social Media.

Abstract

Twitter and other social media sites contain a wealth of information about populations and has been used to track sentiment towards products, measure political attitudes, and study social linguistics. In this talk, we investigate the potential for Twitter and social media to impact public health research. Broadly, we explore a range of applications for which social media may hold relevant data, including disease surveillance, public safety, and drug usage patterns. To uncover these trends, we develop new statistical models that can reveal trends and patterns of interest to public health from vast quantities of data. Our results suggest that social media has broad applicability for public health research.

Bio

Mark Dredze is an Assistant Research Professor in Computer Science at Johns Hopkins University and a research scientist at the Human Language Technology Center of Excellence. He is also affiliated with the Center for Language and Speech Processing and the Center for Population Health Information Technology. His research in natural language processing and machine learning has focused on graphical models, semi-supervised learning, information extraction, large-scale learning, and speech processing. His recent work includes health information applications, including information extraction from social media, biomedical and clinical texts. He obtained his PhD from the University of Pennsylvania in 2009.

J. Sean Humbert, Associate Professor in the Department of Aerospace Engineering at the University of Maryland-College Park, presents Principles of Sensorimotor Control at Insect Scales.  

Abstract
Control architectures for nature’s small, agile flyers exhibit substantial differences from their engineered counterparts. For instance, latency is minimized by neurons that spatially weight across non-orthogonal arrays of sensors as opposed to traditional temporal filtering of small numbers of orthogonally arranged sensors to achieve signal to noise improvements. In this talk I will discuss several of these distinctions in the context of distillable solutions for small-scale engineered systems including: directionality and observability properties of insect sensory structures such as the compound eye; reduced order models of flapping flight dynamics and analysis of high speed videography of insects in free flight to understand controllability properties; implications for efficient coupling of sensing and actuation; and hardware implementations and demonstrations of insect-inspired sensorimotor architectures.

Bio

J. Sean Humbert holds a BS degree in Mechanical Engineering from the University of California, Davis, and MS and PhD degrees in Mechanical Engineering from Caltech. He is currently an Associate Professor in the Aerospace Engineering Department at the University of Maryland, College Park and an affiliate faculty in the Department of Bioengineering and the Institute for Systems Research. Dr. Humbert is the Director of the Autonomous Vehicle Laboratory at the University of Maryland, which conducts research in robotics, estimation, and control theory, with applications to biological sensing and locomotion (insects, aquatics), small scale Micro/Nano Air Vehicles (MAVs), and Autonomous Underwater Vehicles (AUVs). Current efforts include sensing and estimation in small scale biological systems, and flight dynamics and control in insects and robotic aerial microsystems. Dr. Humbert is the recipient of the AIAA National Capital Section Hal Andrews Young Scientist/Engineer Award, and best overall paper awards at the American Control Conference and the AIAA Conference on Guidance, Navigation and Control. He is an ARO Young Investigator and currently Associate Director of the MAST-CTA on Microsystem Mechanics.

Nathan Michael, Research Assistant Professor in the Department of Mechanical Engineering and Applied Mechanics at the University of Pennsylvania, presents Indoor/Outdoor Autonomous Navigation and 3D Exploration with a Quadrotor.

Abstract
In this talk, I will discuss approaches to state estimation and exploration with an autonomous quadrotor in indoor and outdoor environments. Quadrotors, and more generally micro-aerial vehicles, offer 3D mobility and observational capabilities that make them particularly applicable to search-and-rescue scenarios where the vehicle must navigate complex 3D environments for the purposes of mapping and inspection. During search-and-research scenarios in or near buildings, such as in response to an emergency or disaster, it may not be possible for a human to enter the environment and thus the vehicle must be able to access the location and operate autonomously without requiring high-level direction. This talk focuses on the addressing these requirements and will consist of three parts: (1) a brief overview of the necessary prerequisites to enable autonomous navigation with a quadrotor in multi-story indoor environments; (2) a state estimation approach that permits autonomous navigation in mixed indoor and outdoor environments; and (3) an algorithm for exploration in complex three-dimensional environments that is amenable to application on a quadrotor with payload constrained sensing and computational capabilities. I will present the details of the methodologies as well as experimental evaluations in relevant operational environments.

Bio
Nathan Michael recently joined the Robotics Institute at Carnegie Mellon University as an Assistant Research Professor. Prior to this appointment, he was a Research Assistant Professor in the Department of Mechanical Engineering at the University of Pennsylvania. After receiving a PhD from the Department of Mechanical Engineering at the University of Pennsylvania in 2008, he transitioned into a position on the Research Faculty in 2010. Michael's research interests include the topics of estimation and control for ground and aerial robots with extensions to multi-robot systems.

Abstract

In recent years photo and video sharing web sites like Flickr and Youtube have become increasingly popular. Nowadays, every day millions of photos are uploaded. These photos survey the world. Given the scale of data we are facing significant challenges to process them within a short time frame given limited resources. In my talk I will present my work on the highly efficient organization and reconstruction of 3D models from city scale photo collections (millions of images per city) on a single PC in the span of a day as well as my work on the real-time scene reconstruction from video. The approaches address a variety of the current challenges to achieve a concurrent 3D model from these data. For reconstruction from photo collections these challenges are: selecting the data of interest from the noisy datasets, efficient robust camera motion estimation. Shared challenges of photo collection based 3D modeling and 3D reconstruction from video are: high performance stereo estimation from multiple views, as well as image based location recognition for topology detection. In the talk I will discuss the details of our appearance and geometry based image organization method, our efficient stereo technique for determining the scene depths from photo collection images and their depth maps will also be explained during the talk. It allows to perform the scene depth estimation with multiple frames per second from a large set of views with a considerable variation in appearance.

Bio

Jan-Michael Frahm is an Assistant Professor at University of North Carolina at Chapel Hill. He received his PhD in computer vision in 2005 from the Christian-Albrechts University of Kiel, Germany. His Diploma in Computer Science is from the University of Lubeck. Dr. Frahm`s research interests include a variety of computer vision problems. He has worked on structure from motion for single/multi-camera systems for static and dynamic scenes to create 3D models of the scene; real-time multi-view stereo to create a dense scene geometry from camera images; use of camera-sensor systems for 3D scene reconstruction with fusion of multiple orthogonal sensors; improved robust and fast estimation methods from noisy data to compensate for highly noisy measurements in various stages of the reconstruction process; high performance feature tracking for salient image-point motion extraction; and the development of data-parallel algorithms for commodity graphics hardware for efficient 3D reconstruction. He has published over 65 peer reviewed papers in international conferences and journals, and he is Editor in Chief of the Elsevier Journal of Image and Vision Computing.

Robert J. Webster III, Assistant Professor in the Department of Mechanical Engineering at Vanderbilt University, presents Enabling Technologies for More Accurate, Less Invasive Robotic Surgery.

Abstract
This seminar will focus on recent results from the Vanderbilt Medical & Electromechanical Design Laboratory on concentric tube robots, capsule endoscopy, and robotic cochlear implant surgery, together with a brief discussion on fledgling mechatronic STEM education projects for both healthy and disabled students. Concentric tube robots are thin “tentacle-like” robots able to elastically wind around and through delicate anatomy, minimizing damage while accessing challenging surgical sites. Enabling these robots to achieve their potential requires mechanics-based models, image guidance, path planning, and end effector design. The capsule endoscopy portion of the talk will cover swallowable pill-sized robots capable of locomotion, insufflation, and therapy delivery in the GI tract. The talk will also address robot-assisted otologic surgery, which involves bone shaping and drilling in the skull to facilitate cochlear implantation and other inner-ear procedures.

Bio
Robert J. Webster III received his BS in electrical engineering from Clemson University in 2002, and his MS and PhD degrees in mechanical engineering from the Johns Hopkins University in 2004 and 2007. In 2008, he joined the mechanical engineering faculty of Vanderbilt University, where he currently directs the Medical & Electromechanical Design (MED) Laboratory. Dr. Webster's research interests include medical robotics, image-guided surgery, and continuum robotics. He received a National Science Foundation CAREER award in 2011, and an IEEE Volz award for PhD thesis impact on the field of robotics.