TECH010: The Real Robotics Timeline w/ Ken Goldberg (Tech Podcast) - We Study Billionaires Recap

Podcast: We Study Billionaires

Published: 2025-12-24

Duration: 58 minutes

Guests: Ken Goldberg

Summary

Ken Goldberg and Preston discuss the current state of robotics, highlighting the gap between AI language models and robotic dexterity. They explore the challenges in robotic manipulation and the incremental advancements in the field.

What Happened

Ken Goldberg, a prominent robotics researcher, joins Preston to discuss the current landscape of robotics, focusing on the stark contrast between the advancements in AI language models and the slower progress in physical robotics. They highlight the challenges that robotic dexterity faces, such as the complexity of tasks like tying shoelaces, which remain out of reach for current robotic systems due to the lack of tactile sensing and nuanced manipulation capabilities.

While robotic mobility has seen significant advancements with quadrupeds, bipeds, and drones, dexterous manipulation still lags behind. Goldberg points out that, despite the sophisticated design of human-like robotic hands, the control and sensory feedback necessary for nuanced tasks are still lacking. This is evidenced by the fact that simpler robotic grippers often outperform more complex human-like hands in practical applications.

The discussion also covers the significant data gap in robotics compared to AI language models. Language models benefit from a vast repository of training data, equivalent to 100,000 years of human reading, while similar data for robotic manipulation is scarce. Goldberg's company, Ambi Robotics, addresses this by using simple grippers and emphasizing software and control over hardware complexity.

Ambi Robotics has seen success with systems like Dex-Net, which can efficiently sort packages by simulating data and adding noise to train networks on grasping object models. Goldberg details how Dex-Net generated 6 million example grasps using 10,000 object models, demonstrating the potential of combining AI with traditional engineering principles for practical applications.

Another significant point discussed is the role of tactile sensing versus vision in robotic surgery. Goldberg clarifies that robotic surgery often involves a surgeon using a robot as a tool, relying more on visual feedback than tactile sensing, which is a common misconception about the autonomy of surgical robots.

Goldberg also comments on Elon Musk's cost-driven approach in robotics, such as the decision not to use LiDAR in Tesla cars, highlighting how cost constraints can influence technological decisions. He expresses concern over the inflated expectations for humanoid robots and the potential backlash if companies fail to meet these expectations.

Finally, Goldberg reflects on the potential for robots to perform useful household tasks in the future, such as picking up items from the floor. He acknowledges that while simulation works well for certain tasks, complex deformations like tying shoes still pose a challenge, indicating a need for continued development in this area.

Key Insights