Robot training pipeline mapped: Ego Video pre-training, UMI mid-training, teleop post-training
paigeinsf · x · 2026-10-08
A widely shared summary of the robot learning data stack: pre-training runs on egocentric video (Ego Video), mid-training combines Ego and UMI data, and post-training relies on teleoperation plus rollouts.
The thread sparked discussion about who provides a one-stop robotics data and training solution, reflecting the current phase-by-phase division of robot data pipelines.
More from Embodied
- US Treasury issues first outbound investment fine: $200k over $92k Chinese robotics AI deal — pstAsiatech · 2026-10-08
- Palmer Luckey: I Believe in Humanoid Robots, But Won't Build Them Myself — PeterDiamandis · 2026-10-08
- Adding Motion Sense to VLAs With Frozen SAM 2 Lets Robots Catch Moving Bottles — nikhilaravi · 2026-10-08
- Autonomous Trash-Picking Quadruped: Extended Arm Links and a Crouch Trick to Fix Reachability — IsaiahBallah · 2026-10-08
- Microduck robot emotion design open-sourced: 14 emotions pairing motion keyframes with sounds — alexcovo_eth · 2026-10-08
- Safesight, an AI safety layer for existing machines, doubles LOIs and POs to $40M — thisdudelikesAI · 2026-10-08