Category: Seminars

Erik Komendera

3/8/17

Abstract: With the retirement of the Space Shuttle program, the option to lift heavy payloads to orbit has become severely constrained.  Combined with the increasing success and decreasing costs of commercial small- to medium-lift launch vehicles, robotic in-space assembly is becoming attractive for mission concepts such as large space telescopes, assembly and repair facilities, solar electric propulsion tugs, and in situ resource utilization.  Challenges in autonomous assembly include reasoning with uncertainties in the structure, agents, and environment, delegating a large variety of assembly tasks, and making error corrections and adjustments as needed.  For space applications, the design and assembly of each part requires extensive planning, manufacturing, and checkout procedures.  This hinders servicing, and prevents repurposing functional parts on derelict spacecraft.  The advent of practical robotic in-space assembly will mitigate the need for deployment mechanisms and enable assembly using materials delivered by multiple launch vehicles.  This reduction in complexity will lead to simplified common architectures, enabling interchangeable parts, and driving down costs

In recent years, Langley Research Center has developed assembly methods to address some of these challenges by distributing long reach manipulation tasks and precise positioning tasks between specialized agents, employing Simultaneous Localization and Mapping (SLAM) in the assembly workspace, using sequencing algorithms, and detecting and correcting errors.  This talk will describe ongoing research, discuss the results of several recent robotic assembly experiments, and preview the upcoming assembly experiments to be performed under Langley’s “tipping point” partnership with Orbital/ATK.

Bio: Dr. Erik Komendera is a roboticist at NASA Langley Research Center in Hampton, VA. He earned his MS (’12) and PhD (’14) in Computer Science from the University of Colorado, and earned a BSE in Aerospace Engineering (’07) from the University of Michigan.  Dr. Komendera’s current research focuses on autonomous assembly of structures in space, with a special focus on state estimation and machine learning techniques to identify and overcome errors in the assembly process. He currently serves as a task lead on the joint NASA/Orbital ATK Tipping Point project titled “Commercial Infrastructure for Robotic Assembly and Servicing” (CIRAS). In addition, he is Principal Investigator for a LaRC Center Innovation Fund / Internal Research and Development award to investigate machine learning methods for ensuring robust assembly and repair of solar array modules, and is a key member of the “Robotic Assembly of Modular Space Exploration Systems” research incubator effort.

Bennett Wineholt

3/22/17

Recent work to reduce the size and computational requirements of deep neural networks for machine learning has allowed applications including video object recognition and speech recognition to be performed responsively on small robotic systems which are commonly limited by power and payload constraints.  This talk will present an application lifecycle for developing robot behaviors using deep learning techniques as well as describing advances in model compression which make these techniques more performant.

Bio: Bennett Wineholt is a staff member at the Cornell University Center for Advanced Computing supporting faculty needs for computing and consulting services to accelerate discovery.

Patrícia Alves-Oliveria

3/29/17

In this talk Patrícia will present her work on the field of Human-Robot Interaction. Specifically, she will introduce her previous work on the European project EMOTE whose goal was to develop a robotic tutor to support curricular activities in school. Additionally, Patrícia will present her initial work on creativity with robots.

Bio: Patrícia is a PhD student in psychology in an exchange program between Portugal and Cornell University. She is being supervised by Prof. Guy Hoffman and Prof. Ana Paiva (Gaips lab, Portugal) and she is studying how we can use robots to boot creativity in children.

Jesse Goldberg

4/12/17

Reinforcement learning enables animals to learn to select the most rewarding action in a given context. Edward Thorndike posed a simple solution to this problem in his Law of Effect: ‘Responses that produce a satisfying effect in a particular situation become more likely to occur again in that situation, and responses that produce a discomforting effect become less likely to occur again in that situation.’ This idea underlies stimulus-response, reinforcement, and instrumental learning and implementing it requires three pieces of information: (1) the action (response) an animal makes; (2) the context (situation) in which the action is taken; and (3) evaluation of the outcome (effect). In vertebrates, the basal ganglia have been proposed to integrate the three pieces of information required for reinforcement learning: (1) The situation, or current context, is thought to be signaled by a massive projection from the cortex to the striatum, the input layer of the BG; (2) The chosen action is signaled by striatal medium spiny neurons (MSNs) that drive behavior via projections to downstream motor centers; and (3) The evaluation of the outcome is transmitted to the striatum by midbrain DA neurons. These signals underlie a simple ‘three-factor learning rule’: If a cortical input is active (signifying a context), the MSN discharges (driving the action chosen), and an increase in DA subsequently occurs (signifying a good outcome), then the connection strength of the cortical input to the MSN is increased. Overall, by controlling the strength of the corticostriatal synapse, this dopamine-modulated corticostriatal plasticity governs which action will be chosen in a given context, placing DA in the premier position of determining what animals will learn and how they will behave. Here, I will discuss how our recent identification of dopaminergic error signals in birdsong support the potential generality dopamine modulated corticostriatal plasticity in implementing learning in a wide range of behaviors.

Kevin Chen

4/19/17

Abstract: Flapping flight is ubiquitous among agile natural flyers. Taking inspiration from biological flappers, we develop a robot capable of insect-like flight, and then go beyond biological capabilities by demonstrating multi-phase locomotion and impulsive water-air transition. In this talk, I will present our recent research on developing a hybrid aerial-aquatic microrobot and discuss the underlying physics. I will start by describing experimental and computational studies of flapping wing aerodynamics that aim to quantify fluid-wing interactions and ultimately distill scaling rules for robotic design. Comparative studies of fluid-wing interactions in air and water show remarkable similarities, which lead to the development of the first hybrid aerial-aquatic flapping wing robot. In addition to discussing the flapping frequency scaling rule and robot underwater stability, I will describe the challenges and benefits imposed by water surface tension. By developing an impulsive mechanism that utilizes electrochemical reaction, we further demonstrate robot water-air transition. I will conclude by outlining the challenges and opportunities in our current microrobotic research.

Anil Rao

4/26/17

Optimal control concerns systems that evolve in time for which you have partial control of the system and it is desired to optimize a specified performance criterion.   Optimal control problems arise in a variety of applications including engineering, economics, medicine, and epidemiology.

With a few notable exceptions (for example, the brachistochrone problem), virtually no optimal control problems have analytic solutions. Consequently, it is necessary to obtain a solution using numerical methods. Even with modern computers, solving optimal control problems numerically is a challenge because most optimal control problems of interest are nonlinear, high-dimensional, and have complex constraints.  As a result, finding accurate solutions to a general optimal control problem requires the development of sophisticated methods.

This seminar describes a framework for solving constrained optimal control problems.  The key approach described in this seminar is a class of variable-interval (h) variable-order (p) methods, also called hp-adaptive methods. In the hp-adaptive approach, a continuous optimal control problem is approximated as a finite-dimensional nonlinear optimization problem.  This class of hp-adaptive methods are employed using Gaussian quadrature to provide high-accuracy solutions using a significantly lower-dimensional discretization when compared with traditional fixed-order methods.

This seminar will first step through a motivation for the hp-adaptive approach. Recent research done in hp-adaptive mesh refinement techniques will be highlighted along with advances in methods for algorithmic differentiation.  The effectiveness of the approach will be demonstrated using the benchmark Bryson minimum time-to-climb of the F-4 supersonic aircraft.  Specifically, this aircraft flight example will demonstrate the significant improvements in computational efficiency gained by the hp-adaptive approach over previously developed methods.  Furthermore, a low-thrust Earth orbit transfer with eclipsing will be used to demonstrate the capability of the approach on a challenging space flight application.  Finally, future research directions will be discussed.

 BIOGRAPHICAL SKETCH

Anil V. Rao earned a BS in mechanical engineering and and AB in mathematics from Cornell, an MSE in aerospace engineering from the University of Michigan, and  an  MA and PhD from Princeton University. After earning his PhD, Dr. Rao joined the The Aerospace Corporation in Los Angeles abd was subsequently a Senior Member of the Technical Staff at The Charles Stark Draper Laboratory in Cambridge, Mass.  While at Draper, from 2001 to 2006, he was an adjunct faculty in the Department of Aerospace and Mechanical Engineering at Boston University,  where he taught the core undergraduate dynamics course.  Since 2006 he has been in Mechanical and Aerospace Engineering at the University of Florida where he is current an Associate Professor and Erich Farber Faculty Fellow. His research interests include computational methods for optimal control and trajectory optimization, nonlinear optimization, space flight mechanics, orbital mechanics, guidance, and navigation. He has co-authored the textbook Dynamics of Particles and Rigid Bodies: A Systematic Approach (Cambridge University Press, 2006).  He is active in professional societies including the American Institute of Aeronautics and Astronautics, the American Astronautical Society, and the Society for Industrial and Applied Mathematics.  Dr. Rao serves on the editorial board of the Journal of the Astronautical Sciences, the Journal of Optimization Theory and Applications, and the Journal of Spacecraft and Rockets. He is the co-developer of the industrial-strength optimal control software GPOPS-II. His teaching and  research awards include the Department Teacher of the Year at BU (2002 and 2006) and at the University of Florida (2008), the College of Engineering Outstanding Teacher of the Year Award at BU (2004), the Book of the Year Award at Draper Laboratory (2006), the Pramod P. Khargonekar Junior Faculty Award (2012) at the University of Florida. He is an Associate Fellow of the American Institute of Aeronautics and Astronautics.

Thomas Wallin

5/10/17

Conventional robots are composed of rigid components with discrete linkages that promote high precision and controllability; however, these systems require complex sensing and feedback controls and can struggle to perform in uncontrolled conditions.  Soft robots, by comparison, reduce the control complexity and manufacturing cost, while simultaneously allowing new, sophisticated functions.  While earlier generations of soft robots were limited architecturally and functionally, recent advances in materials and additive manufacturing technologies have enabled new and exciting capabilities.   In this talk, I will begin by discussing the essential elements of soft robots, highlighting the pertinent material properties.  Then I will describe the advantages and limitations of the different 3D printing technologies employed in both the indirect and direct fabrication of soft actuators.  For each manufacturing technique, we will discuss the compatible material classes with a focus on actuation and/or sensing mechanisms.