Category: Seminars

Date: 3/31/2022

Speaker: Tianyu Wang

Location: 122 Gates Hall and Zoom

Time: 2:40 p.m.-3:30 p.m.

Abstract:  Autonomous systems operating in unstructured, partially observed, and changing real-world environments need a semantic and logical understanding of the task. Designing a cost function that encodes complex rules by hand is infeasible and it is desirable to learn policies from expert demonstrations. However, it remains challenging to infer the proper cost function from rich semantic information while following the underlying task structure in demonstrations. In this talk, I will first give an overview of an inverse reinforcement learning method that learns to navigate from semantic observations. The efficacy of our method is demonstrated by an autonomous driving task in the CARLA simulator. Second, I will present an automata-based approach to learn the sequential task logic from demonstrations. 

Bio: Tianyu Wang is a Ph.D. candidate in Electrical and Computer Engineering at UC San Diego. His research interests include reinforcement learning, inverse reinforcement learning and autonomous driving. He received an M.S. degree in Electrical and Computer Engineering from UC San Diego and a B.S. degree in Physics from Haverford College.

 

 

 

 

 

Date: 3/24/22

Speaker: Igor Gilitschenski

Location: 122 Gates Hall and Zoom

Time: 2:40 p.m.-3:30 p.m.

Abstract:

In recent years, we have seen an exploding interest in real-world deployment of autonomous systems, such as autonomous drones or vehicles. This interest was sparked by major advances in robot perception, planning, and control. However, robust operation in the “wild” remains a challenging goal. Correct consideration of the broad variety of real-world conditions requires both, better understanding of the learning process and robustifying the deployment of autonomous robots. In this talk, I will discuss several of our recent works in that space. This involves, first, discussing the challenges associated with severe weather conditions. Second, approaches for reducing real-world data requirements for safe navigation. Finally, enabling safe learning for control in interactive settings.

 

Bio: Igor Gilitschenski is an Assistant Professor of Computer Science at the University of Toronto where he leads the Toronto Intelligent Systems Lab. He is also a (part-time) Research Scientist at the Toyota Research Institute. Prior to that, Dr. Gilitschenski was a Research Scientist at MIT’s Computer Science and Artificial Intelligence Lab and the Distributed Robotics Lab (DRL) where he was the technical lead of DRL’s autonomous driving research team. He joined MIT from the Autonomous Systems Lab of ETH Zurich where he worked on robotic perception, particularly localization and mapping. Dr. Gilitschenski obtained his doctorate in Computer Science from the Karlsruhe Institute of Technology and a Diploma in Mathematics from the University of Stuttgart. His research interests involve developing novel robotic perception and decision-making methods for challenging dynamic environments. He is the recipient of several best paper awards including at the American Control Conference, the International Conference of Information Fusion, and the Robotics and Automation Letters.

 

 

 

 

 

Date: 3/17/22

Speaker: Mustafa Mukadam

Location: 122 Gates Hall and Zoom

Time: 2:40 p.m.-3:30 p.m.

Abstract:

Robot perception sits at a unique cross roads between computer vision and robotics. The nature of sensing is ego centric and temporal, and can be from contact-rich interactions. Fusing these signals into representations that enable downstream tasks in real-time is a challenge. In this talk, I will cover some of our recent work in building signed distance fields, human pose and shape estimation, and tactile estimation that provide a recipe for thinking about perception problems with a robotics lens by making optimization and prior models compatible with deep learning.

Bio:

Mustafa Mukadam is a Research Scientist at Meta AI. His work focuses on fundamental and applied research in robotics and machine learning, and structured techniques at their intersection towards practical robot learning. Specifically, his research spans problems from perception to planning for navigation and manipulation. He received a Ph.D. from Georgia Tech where he was part of the Robot Learning Lab and Institute for Robotics and Intelligent Machines. His works have been covered by media outlets like GeekWire, VentureBeat, and TechCrunch, and work on motion planning has received the 2018 IJRR paper of the year award.

 

 

 

 

 

Date: 3/3/22

Speaker: Sanjiban Choudhury

Location: 122 Gates Hall and Zoom

Time: 2:40 p.m.-3:30 p.m.

Abstract:

Advances in machine learning have fueled progress towards deploying real-world robots from assembly lines to self-driving. However, if robots are to truly work alongside humans in the wild, they need to solve fundamental challenges that go beyond collecting large-scale datasets. Robots must continually improve and learn online to adapt to individual human preferences. How do we design robots that both understand and learn from natural human interactions?

In this talk, I will dive into two core challenges. First, I will discuss learning from natural human interactions where we look at the recurring problem of feedback-driven covariate shift. We will tackle this problem from a unified framework of distribution matching. Second, I will discuss learning to predict human intent where we look at the chicken-or-egg problem of planning with learned forecasts. I will present a graph neural network approach that tractably reasons over latent intents of multiple actors in the scene. Finally, we will demonstrate how these methods come together to result in a self-driving product deployed at scale.

Bio: Sanjiban Choudhury is a Research Scientist at Aurora Innovation and soon-to-be Assistant Professor at Cornell University. His research goal is to enable robots to work seamlessly alongside human partners in the wild. To this end, his work focuses on imitation learning, decision making and human-robot interaction. He obtained his Ph.D. in Robotics from Carnegie Mellon University and was a Postdoctoral fellow at the University of Washington. His research has received best paper awards at ICAPS 2019, finalist for IJRR 2018, and AHS 2014, and winner of the 2018 Howard Hughes award. He is a Siebel Scholar, class of 2013.

 

 

 

 

 

Date: 2/24/22

Speaker: De’aira Bryant

 

 

Location: 122 Gates Hall and Zoom

Time: 2:40 p.m.-3:30 p.m.

Abstract: Social robots are robots designed to interact and communicate directly with humans. Yet, many current robots operate in restrictive social environments. In order for these machines to operate effectively in the real world, they must be capable of understanding the many factors that contribute to human social interaction. One such factor is emotional intelligence. Emotional intelligence (EI) allows one to consider the emotional state of another in order to motivate, plan, and achieve their goals. This presentation will first highlight current techniques in artificial intelligence that incorporate aspects of EI in human-robot interactions. This ability is especially important for applications involving children who are often still learning social skills. However, many approaches in artificial EI have not critically considered children in their target populations. The latter portion of this presentation will feature current research projects ethically and responsibly designing EI for robots capable of interacting with children.