1meg
Back to Labs
Emerging

Humanoid Robotics

General-purpose robots are becoming reality. Trained in simulation, deployed in the real world. The missing piece: an app store for physical skills.

Why Humanoids, Why Now?

Human-Shaped World

Our world is designed for human bodies - stairs, doors, tools, vehicles. A humanoid robot can work in any human environment without modification.

AI Breakthrough

Foundation models can now understand natural language commands and translate them to physical actions. The "brain" problem is being solved.

Sim-to-Real Works

Training in simulation then transferring to reality finally works reliably. Robots can learn in millions of parallel simulations overnight.

Hardware Commoditizing

Motors, sensors, and batteries from EVs and drones are making humanoid hardware affordable. Target prices under $20,000.

Key Players

Tesla (Optimus)
Development

Designed for manufacturing tasks. Leverages Tesla's AI and battery expertise. Target: $20,000 price point.

Learn More
Figure
Prototype

Commercial humanoid for logistics and manufacturing. Raised $675M from Bezos, Microsoft, NVIDIA, OpenAI.

Learn More
Boston Dynamics (Atlas)
R&D

Most advanced humanoid mobility. Backflips, parkour, dancing. Now electric (previously hydraulic).

Learn More
Agility Robotics (Digit)
Deployed

Warehouse logistics focus. Already deployed in Amazon facilities for testing.

Learn More
1X Technologies (NEO)
Development

Norwegian company, backed by OpenAI. Focus on home and workplace assistance.

Learn More
Unitree
Commercial

Chinese robotics company. Affordable humanoid (G1 at $16,000) and quadrupeds.

Learn More

Key Concepts

Sim-to-Real Transfer

Training robots in physics simulation (millions of parallel worlds), then transferring learned behaviors to physical robots. Solves the data problem - real-world robot training is slow and dangerous.

Reinforcement Learning

Robot learns by trial and error, receiving rewards for successful actions. In simulation, can attempt billions of trials to master a skill.

Foundation Models for Robotics

Large language models adapted to understand physical tasks. RT-2, PaLM-E, and others can translate natural language commands into robot actions.

Embodied AI

AI that has a physical body and learns through interaction with the real world. The hypothesis: true intelligence requires embodiment.

The Robot Skill Store

Robots are the new smartphones - standardized hardware that runs downloadable software. The opportunity: an app store for physical skills.

Example Skill Packs
Neural network weights trained in simulation, downloaded to any compatible robot
SkillDescriptionDifficulty
Pick and PlaceIdentify objects, grasp them, move to target locationBasic
AssemblyJoin parts together following instructionsIntermediate
Tool UseOperate hand tools, power tools, specialized equipmentAdvanced
NavigationMove through dynamic environments avoiding obstaclesBasic
Human CollaborationWork alongside humans safely, respond to gesturesAdvanced
Dexterous ManipulationFine motor control - threading needles, handling fragile objectsExpert

Timeline

2013

Boston Dynamics Atlas unveiled (DARPA funded)

2019

OpenAI demonstrates Rubik's cube solving with robot hand (sim-to-real)

2022

Tesla reveals Optimus prototype at AI Day

2023

Figure raises $70M, begins humanoid development

2024

Figure 01 demonstrates autonomous warehouse work; raises $675M

2024

Boston Dynamics reveals all-electric Atlas; retires hydraulic version

Physical AI

The next decade will see humanoid robots move from labs to homes. We're tracking every breakthrough.