Speakers

Abstract: Autonomous robots are becoming increasingly prevalent in everyday life, from navigating our roads and sidewalks to assisting in households and warehouses. Yet building robots that can safely and fluently interact with humans in a trusted manner remains an elusive task. Humans are remarkably adept at avoiding collisions seamlessly, even in crowded settings. In this talk, we will discuss how to utilize human interaction data to learn models that describe these interactions and explore techniques to enhance the safety and fluency of robot planning and control. First, I will discuss recent work that combines data-driven techniques with control-theoretic models to learn interpretable models of safe human-robot vehicle interactions. Second, I will discuss recent work on modeling and inferring multi-agent responsibility for avoiding collisions to gain insight into multi-agent vehicle interactions. Third, I will share some recent work on deep generative modeling for robot planning that is capable of incorporating learned interaction insights to produce safe and efficient robot behaviors. 

Abstract: In this talk, I will present continuous and differentiable representations for planning, control, and learning-based robot behaviours. On the spatial side, I will introduce distance-based fields that provide smooth, queryable distance and gradient to static or moving obstacles. These fields integrate naturally with a range of controllers, including Control Barrier Functions, potential fields, Riemannian Motion Policies, and reinforcement learning for whole-body collision avoidance. On the temporal side, I will present sparse continuous parametrisations of robot skills whose derivatives can be shaped through reinforcement learning, enabling adaptation to new conditions while preserving skill shape and smoothness. Together, these spatial and temporal representations provide a differentiable interface for safety to be embedded throughout the perception-action loop rather than enforced as a separate layer. 

Abstract: Agile flight pushes robots to their physical and perceptual limits, where they must make fast decisions under uncertainty with little room for error. This talk looks at how learning-based methods can reach and even exceed human-level agility on real quadrotors, and what we have learned from moving these systems out of simulation and into the real world. Furthermore, I will introduce emerging directions, such as learning continuously in the real world and multi-agent agile flight, and the open challenges they raise for robust, reliable long-term autonomy.

Abstract: TBD

Abstract: TBD