Robot data and foundation models

Facts

Lane

Physical world

Wave

2 of 12 (build order, not a date)

Needs

1 boxes all the way back through hard links

Unlocks

11 boxes downstream through hard links

Leverage

1.7 points · stands in the way of 7 of 53 stages · rank 6 of 57

Cluster

Robots and the physical economy

Evidence as of October 2026

π0.5 tidies homes it never saw in training, though not always at the first try. Video is plentiful, but robots lack data linking joint movements to outcomes: "we don’t have data for that", says Vincent Sitzmann. A 2025 CMU thesis explores three ways to scale it: simulation, cheap teleoperation and human data gathered in the wild.

1. 01Physical Intelligence: π0.5
2. 02Towards Scaling Embodied Data for Robot Learning (CMU)

Around it

Brakes

Loops and off-ramps

• Physical data flywheelreinforcing loop
• Domains make the datareinforcing loop

Openings researchers' judgement, October 2026 scouting

Depth

deep

Players

Skild (~$1.4bn, $14bn+); Physical Intelligence ($600m B, $5.6bn); Generalist ($3bn); NVIDIA GR00T; DeepMind

Wedge

Not another robot brain: real-world evaluation-as-a-service and data curation tooling that scores which trajectories improve a policy, on open GR00T and pi0 weights. startup

1. 01AI Weekly: physical AI H1 2026
2. 02Implicator: PI nears $1bn round
3. 03NVIDIA GR00T