Hello Everyone,
There are many exciting developments in Robotics and AI right now, so let's get together to learn, connect, and create more opportunities for everyone here in Austin.
This event will be virtual, and you can join using the streaming link provided on the event page.
Talk title: Why World Models Are Resurging in Robotics
Description: Simulation has been arguably the most effective way to train robot policies at scale. The ability to be able to parallelize policy learning has enabled many of the recent breakthroughs from dexterous manipulation to full-body humanoid control.
However, the simulation environment is not the real world; expensive techniques such as domain randomization are applied in hopes that the parameters of the real world lie in the distribution of the parameters we train on in simulation.
This talk will explore the rise of world models in robotics and their potential to increase the sample efficiency of robot training. We’ll go into the fundamental concepts behind world models and understand what makes a high-quality world model that can be used to train robot policies that are robust and generalizable to the uncertainties of the real world. Furthermore, we will discuss the current challenges and future directions of world model research, including how we can effectively scale these models to handle increasingly complex tasks and environments.
Bio: Sri Anumakonda is an undergraduate researcher at Carnegie Mellon University, where he is studying Computer Science and Robotics. His research focuses on the intersection of simulation training for dexterous manipulation tasks. He is a member of the Masason Foundation, where he is recognized as one of 34 members in the 2022 cohort, supporting his research in robotics and computer vision. Previously, Sri worked on vision-based learning for autonomous driving, where he developed Generative Adversarial Networks to create synthetic training data for robust vehicle perception and end2end learning for lateral control.
