Physical AI exec predicts dexterous manipulation breakthrough in 2-3 quarters as robots self-code
BradPorter_ · x · 2026-10-28
Brad Porter, speaking at the MIT AI conference, made a specific prediction: fluid generalized physical AI for dexterous manipulation is 2-3 quarters away.
Why he's confident
- His team is seeing clear scaling laws in physical AI research; pre- and post-training recipes work, and they know how to fuse sim and real data, how much data is needed, and the labeling formula that works
- Frontier models now drive continuous self-improvement: an engineer ran a coding agent inside the robot's ROS container with access to lidar and camera topics; with a single prompt, over 1 hour it iteratively wrote code to predict a cart's handlebar position and a controller to attach—and spontaneously added safety checks for control latency
- Rumor has it OpenAI has moved robotics from research to an applied program, considering the science solved
Why still 2-3 quarters: the industry still trains tasks individually; generalization remains elusive. He cites John Carmack's analysis of that gap but believes it's closing fast.
More from Embodied
- Hemlock launches EMG wearables to capture force data robots can't get from video — ycombinator · 2026-10-28
- Stringman: open-source $795 cable robot covers a whole room, with MCP server — GregKamradt · 2026-10-28
- Dyna Robotics aims to put 100 million robots in the world, hiring full-stack talent — JasonMa2020 · 2026-10-28
- From Vietnam soccer kid to Dyna robotics scientist after a breakout training feature — JasonMa2020 · 2026-10-28
- Mitosis Bench: AI agents find only 32.4% of dividing cells in tumor images — yawnxyz · 2026-10-28
- Building an interactive ESP32-S3 badge with Codex — fully offline — OpenAIDevs · 2026-10-28