Speaker	Talk	YouTube link
	Cyrus Neary Assistant Professor University of British Columbia	Engineering AI Systems and AI for Engineering: Compositionality and Physics in Learning Thu, May 30, 2024 · 11:30 AM ET Abstract How can we transform artificial intelligence (AI) and machine learning capabilities into engineering systems for purposes of robotics and autonomy? That is, how can we engineer AI systems within budget constraints, certify them with respect to stakeholder requirements, and ensure that they meet the needs of the end user? Towards answering these questions, my research creates engineering methodologies for the design of AI-driven systems, as well as learning algorithms that leverage the unique characteristics of engineering problems. In this talk, I will begin by presenting compositional approaches to AI system design, which enable independent development and testing of separate learning-enabled modules, and ultimately facilitate the process of reliably deploying their compositions in practice. Then, I will present control-oriented learning algorithms that integrate data with prior physics knowledge, yielding systems that effectively control hardware after mere minutes of data collection and training. Experiments on robot hardware, ranging from ground vehicles to hexacopters, demonstrate the important role that these algorithms play in the fast and reliable transfer of simulation-and-data-driven AI algorithms to their target, real-world operating environments.
	Audrey Sedal and Shiyao Ni Assistant Professor McGill University	Soft Robots Learn to Crawl: Jointly Optimizing Design and Control with Sim-to-Real Transfer Thu, Jun 20, 2024 · 11:30 AM ET Abstract This work provides a complete framework for the simulation, co-optimization, and sim-to-real transfer of the design and control of soft legged robots. The compliance of soft robots provides a form of "mechanical intelligence" -- the ability to passively exhibit behaviors that would otherwise be difficult to program. Exploiting this capacity requires careful consideration of the coupling between mechanical design and control. Co-optimization provides a promising means to generate sophisticated soft robots by reasoning over this coupling. However, the complex nature of soft robot dynamics makes it difficult to provide a simulation environment that is both sufficiently accurate to allow for sim-to-real transfer, while also being fast enough for contemporary co-optimization algorithms. In this work, we show that finite element simulation combined with recent model order reduction techniques provide both the efficiency and the accuracy required to successfully learn effective soft robot design-control pairs that transfer to reality. We propose a reinforcement learning-based framework for co-optimization and demonstrate successful optimization, construction, and zero-shot sim-to-real transfer of several soft crawling robots. Our learned robot outperforms an expert-designed crawling robot, showing that our approach can generate novel, high-performing designs even in well-understood domains. Audrey bio: Audrey Sedal is an Assistant Professor at McGill University and an Associate Member at Mila. She leads the MACRObotics (Morphology, Actuation, Computation for Robotics) group. From 2020-2021, she was a Research Assistant Professor at Toyota Technological Insititute – Chicago. She earned her PhD in Mechanical Engineering at U. Michigan in 2020 and her SB in Mechanical Engineering at MIT in 2015. In 2019, she was named a US Rising Star in Mechanical Engineering. Shiyao bio: Shiyao is an incoming MSc student at McGill University in the MACRObotics group. He earned his BSc (Honours) in Mechanical Engineering also at McGill University, where his thesis evaluated sim-to-real transfer in soft robotics.
	Parker Ewen PhD Student University of Michigan	See to Act, Act to See: The Interdependence of Planning and Perception Wed, Jun 26, 2024 · 1:00 PM ET Abstract Over the past two decades, the field of robotics has made incredible progress towards sophisticated autonomous systems. Unfortunately, much of this progress has focused on robot navigation, and while robots now excel in this domain, they still struggle to interact with their environments and learn from these interactions. This becomes a sticking point, where robots fail at tasks but learn nothing from these failures. In this talk, I discuss how robots can interact with and learn from their environment with three key insights. The first is that robots should be learning about the physical properties of their environment as these properties are often what determine task success. The second is that visual sensing may be a noisy or unreliable sensing modality, and multiple sensing modalities should be used whenever possible. Finally, in order to act effectively, robots should model the uncertainty in sensor outputs and physical property estimates. In particular, I will present on a novel approach for generating probabilistic models of a robot’s environment using tactile and visual sensing modalities and demonstrate that such models enable legged robots to accomplish tasks, such as locomotion, more effectively.
	Giovanni Beltrame Full Professor Polytechnique Montréal	Swarm robotics across scales: a path for practical robot swarms Thu, Jul 4, 2024 · 11:30 AM ET Abstract Swarm robotics relies on many simple robots, governed by local interactions to implement complex behaviours, generally inspired by biological systems. What are the ingredients necessary for the of design artificial swarms for practical applications? This talk will present some of the challenges of practical swarms, identifying in particular the limitations of a leaderless hierarchy, where robots have only a partial view of their mission, as well as ways to circumvent these limitations by exploiting heterogeneous swarms. This talk will use several practical examples, including the control of smart materials, search and rescue missions, and space exploration.
	Gabriel Robert Lead R&D Developer Ubisoft La Forge	Ubisoft la Forge “game bot” team Thu, Jul 18, 2024 Abstract In this presentation we’ll present Ubisoft La Forge (Ubisoft R&D team) with a strong focus on the “Bot” group whose mission is to build ML based game actors simulating players in games. Through multiple concrete in game examples, we will present our work and the challenges we are facing in building bots for AAA games.
	Kevin Sebastian Luck Assistant Professor Vrije Universiteit Amsterdam	Co-Adaptation of Robot Design & Behavior: A Robot Learning Perspective Thu, Jul 25, 2024 · 11:30 AM ET Abstract In this talk I will discuss the problem of co-designing robot morphology and behaviour using robot learning techniques from the fields of deep reinforcement learning. After a short introduction into the problem setting and some remarks about open problems, I will present a few of our works in this area, in particular our recent work on using imitation learning in the co-adaptation setting. The talk will primarily focus on the problem of optimizing robot embodiment and policies, many of the underlying techniques are applicable to the general setting of multi- or cross-embodiment learning across different robot embodiments.