scripod.com

TECH010: The Real Robotics Timeline w/ Ken Goldberg (Tech Podcast)

Shownote

Ken and Preston examine whether robotics has lost its way, echoing Rodney Brooks’ concerns. They dissect the gap between AI language models and physical robotics, focusing on dexterous manipulation, tactile sensing, and visual feedback. IN THIS EPISODE ...

Highlights

In this episode, Ken Goldberg joins Preston Pysh to explore the evolving landscape of robotics and artificial intelligence, questioning whether the field has strayed from its foundational goals. They delve into the disparity between rapid AI advancements and the slower progress in physical robotic capabilities, examining key technical hurdles that continue to challenge researchers and engineers.
02:37
AI systems are now capable of things once thought impossible, even showing creativity.
07:41
Despite online demos, most advanced robotic hands cannot reliably perform fine manipulation tasks.
11:01
Robots cannot replicate the tactile sensing of human fingertips needed for fine manipulation tasks.
12:37
Surgeons can perform complex tasks like appendix removal without tactile sensing, relying solely on vision.
14:45
Elon's refusal to use LIDAR has limited Tesla's driving systems as LIDAR can help in edge cases.
20:22
Robotics lags behind language models due to insufficient data, requiring deliberate generation of manipulation datasets.
27:08
Simple grippers can be more effective than human-like hands in robotics.
34:37
Ambi Robotics has sorted 100 million packages with reliable systems

Chapters

Intro
00:00
Why Ken agrees that robotics may have “lost its way”
02:37
The critical gap between AI language skills and robotic manipulation
03:37
How robot mobility is advancing, but dexterity still lags
04:33
Why tying shoelaces is still too complex for robots
08:15
The role of tactile sensing vs. vision in robotic surgery
12:37
How camera placement in robotic hands affects manipulation
14:45
Why the robot data gap could be 100,000 years behind language models
20:18
Why simpler grippers often outperform human-like robotic hands
25:13
The engineering behind Dex-Net and Ambi Robotics’ success
27:03
How real-world testing exposed unexpected robotic limitations
34:37

Transcript

Preston Pysh: Hey everyone, welcome to this Wednesday's release of Infinite Tech. Today, we're talking AI and robotics, and where there's still key areas of development that need some work. My guest is Ken Goldberg, a leading robotics researcher whose work...