Mechanical and Mechatronics Department Research Seminars

Department Research Seminars:

To receive credit, students must attend the full MME Department Research Seminars listed here and complete the Seminar Attendance Form.

---

Mechanisms and Consequences of Attributing Socialises to Artificial Agents

Speaker: Prof. Emily Corss

ETH Zurich, Switzerland

Theme: Robotics, Social Interaction
--------------------------------------------
Time: 2023-Nov-17, 10:00 am – 11:00 am
Zoom Meeting ID: 966 7094 8019 (link)

Passcode: MME2022

Summary: Understanding how we perceive and interact with others is a core challenge of social cognition research. This challenge is poised to intensify in importance as the ubiquity of artificial intelligence and the presence of robots in society grows. This talk examines how established theories and methods from psychology and neuroscience reveal fundamental aspects of how people perceive, interact with, and form social relationships with robots. Robots provide a resolutely new approach to studying brain and behavioural flexibility manifest by humans during social interaction. As machines, they deliver behaviours that can be perceived as “social”, even though they are artificial agents and, as such, can be programmed to deliver perfectly determined and reproducible sets of actions. This talk highlights work bridging social cognition, neuroscience and robotics, with important implications not only for social robot design, but equally critically, for our understanding of the neurocognitive mechanisms supporting human social behaviour more generally.

Prof. Emily Cross recently joined ETHZ to establish the Professorship for Social Brain Sciences in Zurich, Switzerland. Prior to this, she was based jointly in Sydney, Australia (at the MARCS Institute at Western Sydney University and the Department of Cognitive Science at Macquarie University) and Glasgow, Scotland (at the School for Psychology and Neuroscience and the University of Glasgow). Emily leads a vibrant and diverse research team that uses brain imaging techniques, robots, and complex action training paradigms to explore how experience-dependent plasticity and expertise is manifest across brain and behaviour, and the affective value of performing arts.

 

Please contact the host, Prof. Yue Hu (yue.hu@uwaterloo.ca), if any questions

---

Strategies for Mitigating Fire Performance Problems in New Generation of Concretes and Structures
Speaker: Prof. V.K.R. Kodur
Michigan State University, MI, USA

Theme: Fire
--------------------------------------------
Time: 2023-Nov-16, 3:00 - 4:00pm
In person seminar, Location: E5-2004
Light snacks will be served

Summary: In recent years, the construction industry has shown significant interest in the use of high performance concretes (HPC) in building applications due to the improvements in structural performance, such as high strength and durability, and sustainable solutions that H P C provide as compared to conventional normal-strength concrete (NSC). These high performance concretes, which include concrete types such high strength concrete (HSC), fiber reinforced concrete (FRC), self-consolidated concrete (SCC), and ultra HPC (UHPC), are typically characterized by higher strength, higher sustainability, lower permeability and thus enhanced durability properties. The use of HPC, together with innovative cross-sectional configurations (such as hollow-core slabs, double T beams and deck slabs), and advanced analysis techniques, often lead to slender members in modern structures, especially concrete buildings. Conventional concretes possess good fire resistance properties and hence concrete structures made of NSC exhibit good fire resistance and resiliency properties. However, number of studies have clearly shown that HPC exhibit poor fire resistance properties, as compared to NSC. Specifically, certain HPC types undergo rapid degradation of strength at elevated temperatures and are also susceptible to explosive spalling under severe fire conditions. These poor fire resistance properties of HPC, together with reduced cross sectional sizes of HSC structural members, can lead to lower fire resistance and lower resiliency in HPC structural systems. In the presentation, severe conditions such as high fire intensity, weaker structural configuration, and poorer material characteristics that can be present in modern buildings, with respect to fire performance will be highlighted. The performance problems associated with high performance concretes under fire conditions will be discussed. Examples of innovative strategies for enhancing fire performance and resiliency of HPC structural systems will be presented. Specific guidelines to enhance fire resistance of HPC members, such as the use of bent ties in columns and addition of fibres to HPC to mitigate fire induced spalling, will be discussed. Through case studies it will be demonstrated that by adopting proper strategies, both at material and structural levels, fire resistance and and resiliency of HPC structures can be enhanced.

Dr. Venkatesh Kodur is a University Distinguished Professor and Director of the Centre on Structural Fire Engineering and Diagnostics at Michigan State University. He is an internationally recognized scholar for his contributions in civil and fire engineering fields and his research accomplishments have had major impacts. He has developed fundamental understanding on the behavior of materials and structural systems under extreme fire conditions. The techniques and methodologies resulting from his research is instrumental for minimizing the destructive impact of fire in the built infrastructure, which continues to cause thousands of deaths and billions of dollars of damage each year in the U.S. and around the world. Many of these design approaches and fire resistance solutions have been incorporated in to various construction codes and design standards in the U.S. and around the world. Dr. Kodur has an outstanding record for international research initiatives and collaborations and has collaborated with top researchers and prestigious organizations from about two dozen countries to produce high quality deliverables. He has published results from his research in 500+ peer-reviewed papers in journals and conferences and has given numerous key-note presentations in major international conferences. He is one of the highly cited authors in Civil Engineering and Fire Protection Engineering disciplines, and as per Google Scholar, he has more than 19,800 citations with an "h” index of 79. Dr. Kodur has served in various leadership positions, including as Chairperson (Head) and Associate Chairperson of the Department of Civil and Environment Engineering at MSU, and as Chair of various technical committees of leading professional societies and on editorial boards of leading journals. In recognition of his contributions, he has been elected as Fellow of seven Institutes and Academies; including He has been elected as Fellow of seven Institutes/Academies.

 

Please contact the host, Prof. Yue Hu (yue.hu@uwaterloo.ca), if any questions

---

MOTION OF LARGE PARTICLES IN TURBULENT BOUNDARY LAYERS
Speaker: Prof. Ellen Longmire
University of Minnesota, USA

Theme: Solids, materials, and Fluids

---------------------------------------------------

Time: 2023-Oct-26th, 11:30 am-12:30 am ETD
Zoom Meeting ID: 789 699 0683 (link) Pass Code: MME2023

Abstract: Our work is motivated by the need to understand and predict turbulent particle-laden flows across a range of environmental and industrial applications. We consider a relatively canonical yet challenging experimental flow designed to be accessible to direct numerical simulation. A spherical particle in a turbulent boundary layer undergoes complicated particle-wall and particle-turbulence interactions. Particles with significant diameter are subject to variations in shear and normal forces around their circumference. Wall friction will affect the particle rolling and sliding motions while coherent flow structures can lift the particle away from the wall. To resolve the sphere dynamics in such a flow, 3D tracking experiments were conducted in a water channel facility. The translation and rotation of individual spheres released from rest were tracked over distances of 6d for multiple flow Reynolds numbers and particle-to-fluid density ratios. Simultaneous stereoscopic PIV measurements were acquired in the logarithmic region surrounding the moving spheres. While neutrally buoyant particles typically lift off from the wall upon release, denser particles travel mostly along the wall. The relative contributions of turbulence, wall friction, and mean shear to the resulting particle motions will be discussed for the different cases considered.

Professor Ellen Longmire received an A.B. in physics (1982) from Princeton University and M.S. (1985) and Ph.D. (1991) degrees in mechanical engineering from Stanford University. Since 1990, she has taught and directed research in the Department of Aerospace Engineering and Mechanics at the University of Minnesota. She also served as Associate Dean of the College of Science and Engineering for five years. Professor Longmire uses experimentation and analysis to answer fundamental questions in fluid dynamics that affect industrial, biomedical, and environmental applications. She is a Fellow of the American Physical Society and received the UM Distinguished Women Scholars Award, the McKnight Land-Grant Professorship, and the NSF National Young Investigator Award. She is currently an Editor-in-Chief for Experiments in Fluids. She previously served as Chair of the APS Division of Fluid Dynamics and a member of the US National Committee on Theoretical and Applied Mechanics. In her spare time, she enjoys many outdoor activities as well as cooking and travelling.

 

Please contact the host, Prof. Zhao Pan (zhao.pan@uwaterloo.ca), if any questions.

---

Introducing crack-arrest features in adhesive bonding by manipulating intrinsic and extrinsic dissipations
Speaker: Prof. Gilles Lubineau
KAUST, Saudi Arabia

Theme: Materials

--------------------------------------------
Time: 2023-Oct-11, 9:00 am – 10:00 am
Zoom Meeting ID: 966 7094 8019 (link) Passcode: MME2022

Summary: Reducing emissions of pollutants is a key priority to fight again the climatic changes that we are going to face over the 21st century. More and more pressing environmental constraints are shaping the new technological solutions in the transportation sector, from personal cars up to mass air transport. Automotive and aerospace industries are then seeking new solutions to create lighter structures to reduce fuel consumption or make easier the transition to electrical vehicles. Extreme lightweight structures can today be obtained by using high-performance composites based on continuous fibers and polymeric matrices. These materials that were reserved before to high tech industries, will become even more common tomorrow with the expected cut in the production cost of carbon fibers. Assembling composite parts is, however, still a challenge that often jeopardizes the energy efficiency of structures. Classical joining techniques, such as bolting and riveting, add to the total structural weight and require hole drilling. These result in extra cost due to manufacturing and to extra weight (due to rivets/bolt and locally increased composite thickness to ensure integrity of the substrate). Designers have thought for a long time about replacing the “bolted” solution with integral adhesive bonding. In such joints, the mechanical cohesion between parts is only ensured by an intermediary adhesive layer. This would minimize the additional work and weight needed to realize the assembly. However, integral adhesive bonding is not used for primary structure today, because of its extreme sensitivity to the quality of the substrate preparation that can largely modify the intrinsic performance of the joint. More important for us, the failure of adhesive joints is often unstable: the joint behaves well until the development of a catastrophic crack that would propagate throughout the whole joint, resulting in the loss of the application. In a sense, adhesive-based-design is missing today the “crack arrest” function that is fulfilled by bolts. There are arresting the cracks even in case of premature failure, providing enough time to repair the structure before catastrophic failure.
Our objective is here to introduce new strategies to equip by design adhesive interfaces with crack arrest features. From a practical point of view, we are manipulating the R-curve of the interface by introducing non-local dissipative mechanisms, such as bridging, that will add to the classical cohesive energy of the adhesive.
Pr. Gilles Lubineau is professor of Mechanical Engineering in the Physical Science and Engineering Division and Director of ENERCOMP, a Technology Consortium for Composites in Energy Applications. He is principal investigator of the Laboratory of Mechanics for Energy and Mobility, an integrated environment for composite engineering that he created in 2009 when joining KAUST). Following his “aggregation” in theoretical mechanics, Pr. Lubineau earned a PhD degree in Mechanical Engineering from École Normale Supérieure de Cachan (ENS-Cachan), France. Before joining KAUST, Pr. Lubineau was a faculty member at the École Normale Supérieure of Cachan, and a non-resident Instructor at the École Polytechnique, France. He also served as a visiting researcher at UC-Berkeley and as Interim Dean of the Physical Science and Engineering Division at KAUST. His fields of research include: integrity at short and/or long-term of composite materials and structures, inverse problems for the identification of constitutive parameters, multi-scale coupling technique, nano and/or multifunctional materials. He covers a wide expertise related to most fields of composite materials, with over 200 published papers in journal spanning from material science (Advanced Materials, Macromolecules, etc..), Composites Engineering, all the way to theoretical mechanics (JMPS, CST, Scientific Reports) and applied maths (IJNME, CMAME, etc..). He is also board member for various journals, including the International Journal of Damage Mechanics. Prof. Lubineau is an elected Member of the European Academy of Sciences and Arts.

 

Please contact the host, Prof. Yue Hu (yue.hu@uwaterloo.ca), if any questions

---

Safe and Efficient Robot Learning from Structured Data

Speaker: Dr. Raffaello Camoriano
Politecnico di Torino, Italy
Theme: Robotics, Learning

--------------------------------------------

Time: 2023-Oct-06, 11:30 am – 12:30 pm
In person seminar
Location: E5-2004
Light snacks will be served

Summary: Recently successful machine-learning solutions to longstanding perceptual and decision-making problems focus on accuracy and success-rate metrics to demonstrate superhuman performance. Such endeavors propel the field forward by showing what is achievable, yet with virtually unlimited data, energy, computations, memory, and labeling resources. Meanwhile, real-world domains such as robotics impose a variety of resource budgets on learning systems. This cal ls for the development of more flexible and efficient learning methods in terms of training and prediction time and memory, requiring limited expert supervision. Moreover, robots are required to operate safely in dynamic environments, which requires the ability to operate on structured data (e.g., sequences, manifolds, etc.). In this talk, we will explore a range of solutions for efficient and structured learning. I will introduce recent techniques for trading off time, memory, and accuracy when training large-scale kernel machines. Then, we will see how kernel methods can be extended to enable incremental learning with fixed update complexity, allowing for adaptive learning in real time, e.g., for robot vision, robotic prosthesis control, and dynamics estimation. We will also see how structured prediction enables imitation learning and model learning even in presence of misspecified priors. To conclude, I will present recent work and open challenges on learning for whole-body humanoid robot locomotion control via end-to-end reinforcement learning and efficient trajectory generation via imitation of human motion capture data.
•
Duan A, Batzianoulis I, Camoriano R, Rosasco L, Pucci D, Billard A, A Structured Prediction Approach for Robot Imitation Learning, IJRR (to appear), 2023
•
Viceconte PM, et al. "Adherent: Learning human-like trajectory generators for whole-body control of humanoid robots." IEEE RAL, 2022
•
Ferigo D*, Camoriano R*, et al. "On the emergence of whole-body strategies from humanoid robot push-recovery learning." IEEE RAL, 2021

Dr. Raffaello Camoriano received his B. Sc. in Computer Engineering cum laude in 2011 and his M. Sc. in Robotics Engineering with top grades in 2013 from the University of Genoa (UniGe), Italy. He was awarded a 5-year ISICT/ISSUGE scholarship for the highest-ranking EECS students in Genoa, and a regional government prize for excellent students. He completed his Ph. D. in Bioengineering and Robotics (curriculum in humanoid robotics) at UniGe in 2017, also working at the Italian Institute of Technology (IIT) under the supervision of Prof. Giorgio Metta and Prof. Lorenzo Rosasco. He was awarded the IEEE Computational Intelligence Society Italy Section Chapter s 2017 Best Ph. D. Thesis Award for his Thesis "Large-scale Kernel Methods and Applications to Lifelong Robot Learning". After his Ph. D., Raffaello worked as a PostDoc at IIT on machine learning (efficient and scalable kernel methods, lifelong/incremental learning, and reinforcement learning) and applications to robotics (model learning, locomotion control, and robot vision) in collaboration with Lorenzo Rosasco, Lorenzo Natale, and Daniele Pucci. Raffaello is currently an Assistant Professor at Politecnico di Torino, Italy, and a membr of the ELLIS Unit Turin and the VANDAL Laboratory. His research interests lie at the intersection between the Science and the Engineering of artificial learning systems. In particular, his work is chiefly focused on devising theoretically-grounded Machine Learning methods capable of scaling up to large and structured datasets, adapting to changing conditions in time, and efficiently interacting with their environments to learn faster and generalize better. He also finds Robotics to be a rich source of -and a particularly well-suited testing platform for- challenging open research problems in Machne Learning. His work in this direction mainly spans locomotion control, model learning, and imitation learning, reinforcement learning, and visual object recognition.

Please contact the host, Prof. Yue Hu (yue.hu@uwaterloo.ca), if any questions