Autonomy in Robots

AMRs · humanoids · quadrupeds · drones

Robots share the road-vehicle autonomy stack but not its environment. The more structured the site, the sooner a robot fleet scales.

Where each robot class runs autonomously today, laid out from structured to open environments. Wheeled autonomous mobile robots (AMRs) run at scale in structured warehouses (Amazon, Geek+) and are early in semi-structured yards as yard tractors. Humanoids are in factory pilots in structured sites (Agility Digit, Figure, UBTech). Quadrupeds run inspection fleets in semi-structured industrial sites (Boston Dynamics Spot, ANYbotics ANYmal, Unitree). Drones run docked inspection in semi-structured sites (DJI Dock, Skydio) and delivery services in open airspace (Zipline, Wing, Meituan). Sidewalk bots run on campuses and city sidewalks (Starship, Serve). Autonomy reached scale first in the most structured environments. WHERE EACH ROBOT CLASS RUNS AUTONOMOUSLY TODAY Structured Semi-structured Open / public Wheeled AMRs At scale Amazon, Geek+ Early yard tractors Humanoids Factory pilots Digit, Figure Quadrupeds Inspection Spot, ANYmal Drones Inspection DJI Dock, Skydio Delivery Zipline, Wing Sidewalk bots Campus + city Starship, Serve Structure falls from left to right; autonomy reached scale first on the left.
Amazon robots
1M+
Most mature
Warehouse AMRs
Hardest
Public spaces
Hub
Autonomy

Finding

Robot autonomy scales in the order the operator can control the environment: warehouses first, industrial sites next, sidewalks and open airspace last

Amazon runs more than one million robots across its fulfillment network, nearly all of them wheeled AMRs and drive units on mapped floors with known charging points. Quadrupeds such as Boston Dynamics Spot, ANYbotics ANYmal and Unitree's B2 run routine inspection rounds at plants and substations. Humanoids from Agility Robotics (Digit at GXO), Figure (BMW Spartanburg), UBTech (Walker S at BYD, Zeekr and Foxconn plants) and Tesla (Optimus) are still in factory pilots.

Every class runs the same layers as a robotaxi - perception, localization, planning, control and fallback - but the binding constraint changes with body form. Walking robots must close balance loops at hundreds of hertz, so locomotion is part of the autonomy problem. Drones carry so little energy that endurance, not perception, sets mission length, and a fault is a fall rather than a stop. Sidewalk bots move slowly but meet the densest mix of pedestrians, curbs and local rules of any autonomous machine.

China leads on volume in several classes: Unitree and DEEP Robotics ship more quadrupeds than Western makers, Meituan runs drone delivery routes in Shenzhen, and UBTech, Unitree, AgiBot and Fourier lead humanoid shipments. In the US, the FAA's proposed Part 108 rule for beyond-visual-line-of-sight flight would turn drone delivery from waivers into a standard approval path. Tesla builds Optimus at Giga Texas (AustinIO).

Robot classes

Autonomy by robot class

Maturity is an ElectronsX assessment.

Autonomy by robot class
ClassTypical environmentBinding constraintMaturityLeading players
Wheeled AMRsWarehouses, factories, yardsFleet traffic, dock contentionAt scaleAmazon, Geek+, Hai Robotics, Locus, Symbotic, Quicktron
HumanoidsFactories, warehousesBalance, manipulation, generalizationPilotsTesla, Figure, Agility, Apptronik, UBTech, Unitree, AgiBot
QuadrupedsPlants, substations, tunnels, rough terrainBattery endurance, terrainCommercialBoston Dynamics, ANYbotics, Unitree, DEEP Robotics
Drones and UAVsAirspace over sites and delivery corridorsEnergy, airspace rulesCommercialDJI, Skydio, Zipline, Wing, Meituan, EHang
Sidewalk delivery botsCampuses, neighborhoodsPedestrians, curbs, local rulesEarly serviceStarship, Serve Robotics, Coco, Neolix (road)

Variables

What makes one deployment harder than another

Robot autonomy variables
VariableEasier caseHarder case
Environment structureWarehouse robot on a known mapHumanoid in a mixed-use building
LocomotionWheels on flat floorsLegs on stairs and loose ground; rotors in wind
Human densityFenced industrial yardSidewalk at rush hour
Energy marginDock-connected robot that charges between tasksSmall drone with minutes of endurance
Task scopeOne repeatable transport or inspection jobGeneral-purpose manipulation

Deployment

Where robot autonomy wins first

Robot autonomy deployment order
PatternEarly winnerHarder frontier
Structured sites deploy firstWarehouse and factory AMRsGeneral-purpose robots in public spaces
Energy sets mission timeDock-based and return-to-base robotsLong-endurance legged and aerial robots
Orchestration matters at scaleSingle-vendor fleets under one controllerMixed fleets from several vendors
Generalization is hardSingle-task robotsHumanoids across many tasks

Sources

Sources and scope

Scope
Autonomy in Robots
Sources
Amazon; Agility Robotics; Figure; UBTech; Boston Dynamics; ANYbotics; Unitree; FAA (Part 108 proposal, 2025); ElectronsX assessment
Updated
October 2026