Note: Event details may change. Please refer to the University of Toronto Robotics Institute’s events page for the most current information.

Speaker:

Graham Schelle

Talk title:

Physical AI from Cloud to Robot: An AMD Perspective

Date: Wednesday, September 23, 2026

Time: 3-4 p.m.

Location: Myhal Centre for Engineering Innovation & Entrepreneurship, Room 580

Abstract:

Building intelligent robots requires more than deploying a model to hardware. Modern robotics workflows span cloud-based training, simulation, fine-tuning, and efficient on-device inference, all supported by increasingly capable development tools. This talk explores how these pieces fit together targeting the newly launched AMD Kria AI Developer Platform.

From a collection of AI trends, we will highlight the latest model architectures, fine-tuning techniques, and hardware acceleration that enable efficient deployment of embodied AI applications.  The talk will also examine the growing role of AI coding agents in robotics, showing how they can accelerate software development and complement existing physical AI workflows before a robot is ever deployed.

At AMD, we have the benefit of working with customers and academic collaborators on all these aspects of robotics.  With that perspective in mind, while we are sharing our experiences, we also hope to learn from the University of Toronto community on how we might engage more.

Bio:

Graham Schelle is a fellow at AMD within the Research and Advanced Development Group (RAD) based in Colorado, USA.  He leads a team within AMD Research looking at Physical AI strategies across cloud and edge full-stack deployments.  Specifically, with the open-source Ryzers effort, the team has mapped a variety of Robotics frameworks onto AI PCs, reimagining client devices as embedded platforms for Physical AI workloads.  He is also a lead maintainer of the PYNQ and the RFSoC-PYNQ open-source projects for adaptive SoCs at AMD.