What is this?
AI robotics is where large models meet the physical world. The same technology that understands language now watches, reasons and moves — and the race to a useful humanoid robot has become one of the most funded frontiers in tech, from factories to (eventually) homes.
Key tools & players
- Google DeepMind — Gemini Robotics, the leading "robot brain" line
- Figure — Helix model, humanoids piloted in logistics
- Tesla — Optimus, betting on car-factory scale
- Unitree — China's affordable humanoids and robot dogs
- Boston Dynamics, 1X, Agility, NVIDIA (GR00T, simulation)
Milestones
- 2023 — DeepMind RT-2: web knowledge transfers to robot actions
- 2024 — Figure + OpenAI demo a conversing, reasoning humanoid
- 2025 — Gemini Robotics brings vision-language-action models to many robot bodies; Unitree drops humanoid prices under $6k
- 2026 — Gemini Robotics 2 controls a humanoid's whole body — walking, balance and hands — with one model (our coverage)
- Aug 2026 — Nvidia opens Alpamayo 2 Super for commercial use: a 34B driving VLA under a licence that allows selling derivatives
- Aug 2026 — Unitree lists on Shanghai's STAR Market and opens 629% above its 150.8-yuan offer price: the first humanoid robot maker with a daily public share price, and the only one shipping at a profit
- Aug 2026 — Pony.ai and Uber commit to over 2,000 robotaxis across five European cities; Pony.ai says it has reached city-wide breakeven unit economics in China
- Aug 2026 — Generalist AI's GEN-1.5 learns a new physical task from a single 3–12 second demo with no retraining: the demonstration becomes the prompt (our coverage)
Mini glossary
- VLA: vision-language-action model — sees, understands, then moves
- Teleoperation: humans remotely driving robots to collect training data
- Sim2real: training in simulation, transferring skills to real machines