NYU Professor Named Top Young Innovator for Teaching Robots with Trial and Error

MIT Technology Review released its annual list of top innovators under 35, including NYU professor Lerrel Pinto.
Pinto teaches robots household tasks by letting them fail and learn.
He set up a model home with toy appliances for the robots to practice with.
The technique is called reinforcement learning - the robot tries something, judges its performance, and tries again.
Pinto published a paper "Teach a Robot to FISH" with examples of how tasks were trained.

hackaday.com

Relevant topic timeline:

8/22/2023

A deep learning technique to improve how robots grasp objects

Researchers at the University of Bonn's Autonomous Intelligent Systems research group have developed a deep reinforcement learning pipeline to enhance a robotic arm's ability to manipulate objects in ways that better support practical use, potentially contributing to the development of more effective robotic assistants.

9/19/2023

Toyota Uses Kindergarten to Teach Robots Skills By Showing, Not Coding

Toyota Research Institute (TRI) has used generative AI to teach robots how to perform individual tasks needed to make breakfast by giving them a sense of touch and plugging them into an AI model, enabling them to "learn" and carry out tasks in a matter of hours.

9/20/2023

Toyota's New AI Allows Robots to Master Complex Kitchen Tasks Just by Watching Humans

Toyota has developed a new learning approach called Diffusion Policy, in partnership with Columbia Engineering and MIT, which enables robotic AIs to learn complex physical tasks by observing human demonstrations, allowing them to quickly acquire new skills and perform a wide variety of tasks with flexibility and adaptability.

10/7/2023

Disney Unveils Expressive Walking Robot That Uses AI to Bring Animations to Life

Disney Research has developed a robotic character that uses reinforcement learning to create expressive motions, making it seem uniquely alive and emotive. The system combines the talents of animators and roboticists to develop highly expressive motions that can be trained in just a few hours, reducing the time required to develop a new robotic character. Disney plans to use this technique to develop more physical robotic characters in the future.

10/16/2023

AI Learns Naval Defense Tactics Through Trial and Error

Reinforcement Learning is a computational approach that enables machines to learn from their environment through trial and error, similar to the way young children learn, and has potential applications in various fields including warfare.

Topics

Posted 9/17/2023, 11:00:00 AM