About the show
Researcher's Guide to Physical AI
A weekly briefing on Physical AI: the papers, robots and deals that change what gets built, explained from the ground up and then taken apart properly.
- 14 episodes
- ~14 minutes each
- Weekly new every weekend
- Since August 8, 2026
What it covers
Each topic opens a search of every episode.
- Vision-language-action models
- World models
- Humanoids
- Manipulation
- Locomotion
- Simulation
- Benchmarks
- Robot data
- Autonomous driving
- Safety
- Funding and industry
Who it's for
Researchers and engineers who want the week compressed without losing the technical detail, and anyone curious enough to follow along. Every episode builds the background before going deep, so you don't need to know the field already.
A written deep dive with every episode
The audio is the briefing. The article on each episode page is the reference: full method details, benchmark numbers, limitations, and links to every paper discussed. Start with the latest.
How the audio is made
Shubham narrates the show. Some sections may be voiced by his AI persona, a model trained on his own voice. Either way the words, the research and the editorial calls are his.
Corrections welcome
If a paper of yours was covered and the reading was wrong, say so: the correction is read out in the next episode. Suggestions and papers worth covering are welcome too, at shubham@shrivastava.ai.
Your host
Shubham Shrivastava
Shubham Shrivastava leads AI at Kodiak, where his team builds the world foundation models and camera-lidar-radar perception behind the first 24/7 driverless freight fleet, real trucks, paying freight, no safety driver. He led perception at Ford's autonomous vehicle program before that. He reads the research here the way he reads it at work: not whether a result is impressive, but whether it survives contact with the road.