Marketplace Tech How to train your humanoid robot
Nov 13, 2025
Ayanna Howard, a leading roboticist and dean at Ohio State University, delves into the intriguing world of humanoid robots. She explains how they learn by mimicking human actions through extensive observation. Howard highlights the rise of gig economies that crowdsource data collection to improve robot training. She believes the future of humanoid assistants lies in elder care rather than luxury, emphasizing the need for ethical programming and social skills. Ultimately, she envisions robots as capable companions that can engage meaningfully with humans.
AI Snips
Chapters
Transcript
Episode notes
Learning By Demonstration Requires Lots Of Data
- Robots learn tasks by observing human steps and translating them into robot actions.
- Successful generalization requires huge amounts of varied demonstration data to form reliable patterns.
Robotics Adds Motor And Force Complexity
- Robotic learning resembles generative AI pattern matching plus motor actuation probabilities.
- But many companies still lack sufficient tactile and force-sensing data for robust behavior.
Gig Workers Generate Training Demonstrations
- Companies recruit gig workers to perform tasks repeatedly while wearing cameras and sensors.
- Workers crack eggs, stir, and repeat many variants to capture real-world variability for training robots.