137AI > Agents > Physical Agents


Physical AI Agents


Physical agents are autonomous AI systems that move through and act on the physical world. They drive on public roads, walk through warehouses and homes, deliver packages along sidewalks, haul freight between cities, and fly through controlled and uncontrolled airspace. What sets them apart from other AI systems is that compromise or failure produces motion, contact, transport, or manipulation in real space, with consequences that information-only systems cannot produce. The category spans seven entity types operating under different regulatory regimes and in different environments, each covered on its own page.


The Seven Physical Agent Types

Agent Type Operating Environment Primary Regulatory Frame
Robotaxis & Autonomous Vehicles Public roads alongside human-driven vehicles NHTSA, state DMVs, transportation safety law
Humanoid Robots Warehouses, factories, retail, hospitality, and emerging consumer settings OSHA, industrial machinery safety, ISO 10218, robotics safety standards
Delivery & Mobile Robots Sidewalks, building interiors, residential and commercial neighborhoods Municipal sidewalk and right-of-way ordinances, state-by-state mobile robot laws
Industrial Mobile Robots & Cobots Manufacturing floors, logistics facilities, fulfillment centers OSHA, ANSI R15.08, ISO 3691-4, industrial safety codes
Autonomous Trucks & Platoons Interstate highways, freight corridors, ports, distribution yards FMCSA, NHTSA, state autonomous trucking statutes
Drones & UAS Controlled and uncontrolled airspace, urban and rural environments FAA Part 107, Part 135, remote ID, airspace authorization regimes
Multi-Agent Fleets & Swarms Coordinated operations spanning multiple physical agents across one or more categories above Composite of underlying agent regimes plus emerging fleet coordination policy

What Makes Physical Agents a Distinct Category

Three properties separate physical agents from other AI systems and from each other.

The first is force in the world. A robotaxi can collide. A humanoid can lift, push, or strike. A delivery robot can block a sidewalk or be tipped into traffic. The risk surface includes injury and property damage in ways that software-only systems do not produce.

The second is public-space presence. Physical agents share environments designed for humans and operate among people who did not opt into the interaction. Bystanders, pedestrians, neighbors, and other road users are exposed regardless of consent.

The third is remote orchestration at scale. Single operators or compromised orchestration layers can coordinate thousands of physical agents at once, turning what would be isolated incidents into fleet-wide events.


Where Existing Rules Fall Short

Most regulation governing physical agents was written for non-autonomous predecessors. Vehicle safety law assumed a human driver as the primary safety actor. Industrial machinery safety assumed a stationary or fixed-path machine in a controlled environment. Aviation regulation assumed a piloted aircraft with a licensed operator. Each assumption weakens or fails when the agent is autonomous, mobile, dexterous, and remotely orchestrated.

The result is a patchwork in which autonomous variants of established categories operate under rules that do not quite fit, while regulators, insurers, and operators work out what should replace them. Individual entity pages document the specific gaps for each agent type.


Related Coverage

Convenience as Attack Surface | Personal & Ambient Agents | Software Agents | Human Risks