The Book of Ingenious Devices
Al-Jazari describes 50 mechanical devices, including programmable automata, in his treatise.
Source ↗The Robotics Atlas · updated September 2026
Eight chapters on the machines that are learning to work alongside us: where they came from, how they are built, who is building them now and the rules that govern them. Every fact links to its source.
01 · By the numbers
02 · Timeline
From a thirteenth-century treatise on automata to the first humanoids rolling off production lines.
The Jaquet-Droz automata
The Writer, the Draughtsman and the Musician: clockwork figures that still work today in Neuchâtel.
Source ↗The word “robot”
Karel Čapek’s play R.U.R. premieres. Čapek credited his brother Josef with coining the word.
Source ↗The Three Laws
Isaac Asimov states his Three Laws of Robotics in the short story “Runaround”.
Source ↗Unimate goes to work
The first industrial robot is installed on a General Motors line in Ewing Township, New Jersey.
Source ↗Shakey thinks
SRI begins Shakey, the first mobile robot able to reason about its own actions (to 1972).
Source ↗The first cash machine
Barclays opens the world’s first cash machine in Enfield, London. Finance meets automation.
Source ↗The Stanford Arm
Victor Scheinman builds an all-electric, six-axis, computer-controlled arm.
Source ↗WABOT-1
Waseda University completes what is described as the first full-scale humanoid robot.
Source ↗Honda P2
The first self-contained biped humanoid walks, carrying its own power and computer.
Source ↗ASIMO and da Vinci
Honda introduces ASIMO; the FDA clears the da Vinci surgical system in July.
Source ↗Roomba
iRobot launches the robot vacuum that puts autonomy in millions of homes.
Source ↗Stanley wins
Stanford’s autonomous car wins the DARPA Grand Challenge a year after no vehicle finished.
Source ↗Robots in the bank
MUFG trials NAO robots in Tokyo branches; DARPA’s Robotics Challenge finals are won by KAIST’s DRC-HUBO.
Source ↗Spot for sale
Boston Dynamics puts its quadruped Spot on commercial sale in the US at $74,500.
Source ↗Vision-language-action
Google DeepMind’s RT-2 turns vision and language models into robot actions.
Source ↗Electric Atlas
Boston Dynamics retires hydraulic Atlas and unveils an all-electric humanoid.
Source ↗Humanoids ship
Figure 03 is unveiled and UBTech begins mass delivery of Walker S2.
Source ↗The humanoid year
AgiBot builds its 10,000th humanoid, Unitree lists in Shanghai, and IFR counts 5 million factory robots.
Source ↗DEV 0.0 pilots
Devdoot’s target for its first embodied pilots with robotics hardware partners.
Explore →Drag, swipe or use the arrow keys
03 · Anatomy
Tap a point on the figure. Each system is a field of engineering in its own right.
Anatomy of a humanoid
Brain
Modern humanoids pair fast low-level controllers with large AI models. Vision-language-action models turn what the robot sees and is told into motor commands.
VLA modelOnboard GPU/NPUTask planningFleet learning
04 · How robots learn
- 01
Teleoperate
Humans drive the robot remotely, often in VR, and every movement is recorded as training data.
- 02
Imitate
Models learn to copy those demonstrations, mapping what the robot sees to what it should do next.
- 03
Simulate
Millions of practice runs happen in physics simulators, far faster than real time.
- 04
Transfer
Sim-to-real techniques close the gap between the simulator and messy real floors, light and friction.
- 05
Deploy and share
Robots work, log what happens and share what they learn across the fleet.
05 · Humanoids 2026
The leading humanoid programs and their latest public status, with the date we checked. This field moves monthly.
12 programs
USAJul 2026
Optimus
Tesla · Factory
Production announced for Fremont from mid-2026, starting slowly. No external sales yet.
Source ↗USAJun 2026
Figure 03
Figure AI · Factory · Home
Deployed at BMW Spartanburg for logistics sequencing after an 11-month Figure 02 pilot.
Source ↗USA2026
Digit
Agility Robotics · Logistics
Working with GXO, Schaeffler, Toyota Canada and Mercado Libre; going public via SPAC at a $2.5B valuation.
Source ↗USA / KoreaJan 2026
Atlas (electric)
Boston Dynamics · Factory
Production version shown at CES 2026; first units to Hyundai and Google DeepMind, general availability early 2027.
Source ↗ChinaAug 2026
G1 · H1 · R1
Unitree · Research · Consumer
Listed on Shanghai’s STAR Market; shipped over 5,500 humanoids in 2025. R1 launched at $5,900.
Source ↗ChinaMar 2026
Multiple models
AgiBot · Factory · Service
Rolled out its 10,000th humanoid, three months after its 5,000th.
Source ↗Norway / USAApr 2026
NEO
1X · Home
Preorders at $20,000 or $499 a month; Hayward, CA factory opened with US home deliveries planned for late 2026.
Source ↗USAJun 2026
Apollo
Apptronik · Factory · Logistics
Raised over $935M; Jabil manufactures, Mercedes-Benz is a customer, fleets run with Google DeepMind.
Source ↗ChinaNov 2025
Walker S2
UBTech · Factory
Mass production and delivery began with Walker orders above ¥800M from carmakers and logistics firms.
Source ↗ChinaSep 2026
IRON
Xpeng · Factory · Retail
First robot off a new automated line; mass production planned by end of 2026, deliveries in 2027.
Source ↗ChinaJan 2026
GR-3
Fourier · Care
A care- and companion-focused humanoid shown at CES 2026.
Source ↗Canada2026
Phoenix
Sanctuary AI · Hands · Physical AI
Refocusing on Physical AI software and dexterous hands under a new CEO.
Source ↗06 · Robots in finance
- 1967
The first cash machine
Barclays opens the world’s first cash machine in Enfield. Actor Reg Varney makes the first withdrawal, with a £10 limit.
Source ↗ - 2015
NAO greets customers at MUFG
Bank of Tokyo-Mitsubishi UFJ trials small NAO humanoids to welcome branch visitors in Japan.
Source ↗ - 2018
Pepper on Fifth Avenue
HSBC deploys SoftBank’s Pepper at its New York flagship, the robot’s first US retail-bank role.
Source ↗ - Vaults
The automated vault
G+D built a fully automated cash vault for the National Bank of Rwanda, with robotic handling and storage.
Source ↗ - 2024–26
Robots in the data center
Google tests robots for rack and drive handling; Meta trials dual-arm cabling robots. The infrastructure behind finance is getting hands.
Source ↗ - Next
Where DEV 0.0 fits
Not a greeter. A governed worker for concierge, custody and infrastructure, whose every action is on the record.
Meet DEV 0.0 →
07 · Safety & rules
Asimov’s Three Laws
A robot may not injure a human; must obey orders unless that causes harm; must protect itself unless that conflicts with the first two. Influential, but fiction, not engineering.
Source ↗ISO 10218-1/-2:2025
The core safety standard for industrial robots and their integration, revised in 2025 to absorb collaborative-robot rules and add cybersecurity.
Source ↗ISO/TS 15066
Set force and pressure limits for robots working beside people; its content is now folded into ISO 10218:2025.
Source ↗ISO 13482
Safety requirements for robots that help people directly. Under revision.
Source ↗ISO 25785-1
The first standard aimed at dynamically stable industrial mobile robots, including bipeds. Still a committee draft.
Source ↗EU Machinery Regulation
Regulation (EU) 2023/1230 replaces the Machinery Directive and covers machines with self-evolving behaviour.
Source ↗EU AI Act
The 2026 omnibus pushed high-risk obligations to Dec 2027 (standalone) and Aug 2028 (embedded), with machinery to get its own AI rules.
Source ↗A summary for general reading, not legal advice.
08 · Glossary
- Actuator
- The motor-and-gear unit that moves a joint.
- Backdrivability
- How easily a joint can be pushed by outside force; important for safe contact with people.
- Cobot
- A collaborative robot designed to work near people with limited speed and force.
- Degrees of freedom (DoF)
- The number of independent ways a robot can move. A human arm has about seven.
- Devdoot Ledger
- Devdoot’s on-chain record of every machine action, built so outcomes can be replayed and audited.
- Dexterous manipulation
- Handling objects with fine, multi-finger control rather than a simple gripper.
- Doot
- A Devdoot machine worker with a defined role, permissions and an audit trail.
- E-stop
- An emergency stop that cuts motion immediately; required on industrial robots.
- Embodied AI
- AI that perceives and acts through a physical body instead of only on a screen.
- End effector
- Whatever is at the end of the arm: a hand, gripper, suction cup or tool.
- Fleet learning
- Robots share data and model updates so each one benefits from all the others’ experience.
- Force control
- Controlling how hard a robot pushes, not just where it moves.
- Harmonic drive
- A compact, zero-backlash gear often used in robot joints.
- Humanoid
- A robot with a broadly human body plan: head, torso, two arms and usually two legs.
- Imitation learning
- Training a robot by showing it demonstrations of a task.
- IMU
- Inertial measurement unit: accelerometers and gyroscopes that sense motion and tilt.
- Inverse kinematics
- Working out joint angles needed to put a hand or foot in a given place.
- Lidar
- A laser sensor that measures distance to build a 3D map.
- Locomotion policy
- The learned controller that decides how a legged robot steps and balances.
- Payload
- The weight a robot can carry or lift.
- Physical AI
- The industry term for AI models built to act in the physical world.
- Reinforcement learning
- Learning by trial and error with rewards, often in simulation.
- Series elastic actuator
- An actuator with a spring in line, making motion softer and forces measurable.
- Sim-to-real
- Transferring skills learned in simulation to a physical robot.
- SLAM
- Simultaneous localization and mapping: building a map while tracking your place in it.
- Tactile sensing
- Sensors that let a robot feel pressure, texture and slip.
- Teleoperation
- A human controlling a robot remotely, for work or for collecting training data.
- VLA model
- Vision-language-action model: turns images and instructions directly into robot actions.
- Whole-body control
- Coordinating every joint at once so a robot can balance while it reaches and works.
- Zero-moment point
- A classic balance concept: the point on the ground where tipping forces cancel out.