ESP Biography

A hot dog bun moves slightly, causing a robot chef to deliver an unfinished meal to a customer. A dolphin trainer wears boots with a different color, and suddenly all of his dolphins refuse to listen to him. These failures are actually two sides of the same coin: situations with unexpected circumstances. This is one example of how the core problems in robot-learning are more than technical challenges--they poke at something deeper in how we learn as humans and animals. If we want to make machines that learn like us, we must understand our own mechanisms. In this interactive class, you will become the trainer and try to train your classmates (and the instructor)! We will use these exercises, as well as stories from the wet n’ wild world of orca trainers, to formulate the key challenges of robot learning and how we might solve them. Some examples include exploitation vs. exploration, distribution shift, and causality. These challenges continue to influence cutting-edge robotics research at Stanford. Ultimately, after this class, my hope is that you will have a newfound appreciation for the bidirectionality of engineering. Nature inspires new designs, but these new designs can also lead to a better understanding of nature.

Why is it that machine learning models can beat chess grandmasters, and yet not a single robot can reliably pour water between two arbitrary cups? The challenges behind machine learning can be pretty counterintuitive. As it turns out, we can learn a lot from what animal trainers struggle with! In this interactive class, you will become the trainer and try to train your classmates (and the instructor)! We will use these exercises, as well as stories from the wet n’ wild world of orca trainers, to formulate the key challenges of machine learning. Some examples include exploitation vs. exploration, distribution shift, and causality. These challenges continue to influence cutting-edge robotics research at Stanford. Ultimately, after this class, my hope is that you will have a newfound appreciation for the bidirectionality of engineering. Nature inspires new designs, but these new designs can also lead to a better understanding of nature.

Why is it that machine learning models can beat chess grandmasters, and yet not a single robot can reliably pour water between two arbitrary cups? The challenges behind machine learning can be pretty counterintuitive. As it turns out, we can learn a lot from what animal trainers struggle with! In this interactive class, you will become the trainer and try to train your classmates (and the instructor)! We will use these exercises, as well as stories from the wet n’ wild world of orca trainers, to formulate the key challenges of machine learning. Some examples include exploitation vs. exploration, distribution shift, and causality. These challenges continue to influence cutting-edge robotics research at Stanford. Ultimately, after this class, my hope is that you will have a newfound appreciation for the bidirectionality of engineering. Nature inspires new designs, but these new designs can also lead to a better understanding of nature.

How do you train a dolphin or a killer whale? And what can AI learn from it? Take a dive into the surprising connections between animal training and cutting-edge artificial intelligence algorithms. From Stanford’s robot labs to the wet n’ wild world of orca trainers, hear stories of frustration, innovation, and triumph. In this class you will learn the basics of animal training and reinforcement learning. We will look at how these two seemingly disjoint fields share many of the same insights. To round it all out, we will look at how behavior techniques can be used to improve your interpersonal skills!

How do you train a dolphin or a killer whale? And what can AI learn from it? Take a dive into the surprising connections between animal training and unsupervised artificial intelligence algorithms. In this class you will learn the basics of animal training and reinforcement learning. We will look at how these two seemingly disjoint fields share many of the same insights. Topics covered include operant conditioning, LRS (least reinforcing scenario), behavior cloning, soft actor-critic, and more!