Marketplace All-in-One How to train your humanoid robot
Nov 13, 2025
Ayanna Howard, a leading roboticist and dean at Ohio State University's College of Engineering, dives into the complexities of training humanoid robots. She discusses how these robots learn by observing human actions, facing challenges with data translation and generalization. The conversation highlights the gig economy's role in providing training data and the timeline for practical humanoid assistants. Ayanna also explores the potential of robots as caregivers and the importance of social interaction for acceptance beyond just completing chores.
AI Snips
Chapters
Transcript
Episode notes
How Robots Learn From Demonstration
- Robots learn tasks by observing human steps and mapping them to robot actions via translation of joints and forces.
- Successful generalization needs huge datasets so robots can extrapolate from varied demonstrations.
Robotics Adds Force To Pattern Recognition
- Robotics mirrors generative AI pattern recognition but adds motor actuation and force as critical dimensions.
- Lack of sufficient tactile and force data limits robots' ability to generalize across contexts.
Collect Diverse, Repeated Human Demonstrations
- Recruit gig workers to perform repeated task demonstrations while wearing cameras and sensors to collect varied data.
- Supply diverse examples (different bowls, forces, conditions) so models can generalize to real-world variations.