Physical AI

Definition

Physical AI refers to AI systems embodied in machines that sense their environment, reason about it and take physical actions: robots, autonomous vehicles, drones and industrial equipment driven by foundation models.

How it works

1. 01Sensors (cameras, lidar, force, audio) stream data into perception models.
2. 02Vision-language-action models map perception and instructions to motor commands.
3. 03Simulation and synthetic data generate the scale of training experience that real-world trials cannot.
4. 04Safety layers constrain actions within physical and regulatory limits.
5. 05Fleet learning aggregates experience across deployed machines.

Latest developments

• 2026-08Humanoid pilots in logistics report early task-level metrics.
• 2026-05General-purpose robot foundation models released by multiple labs and chipmakers.
• 2026-01Simulation platforms position as the training ground for embodied AI.

Use cases

• Warehouse picking and material handling
• Manufacturing assembly and inspection
• Autonomous logistics and delivery
• Healthcare and eldercare support
• Agriculture and infrastructure inspection

Market

A capital-intensive market spanning humanoids, industrial arms, mobile robots and the chips, sensors and simulation tooling beneath them. Revenue today is concentrated in industrial automation; humanoids remain pilot-stage.

Risks

• Safety in shared human environments
• Hype outrunning deployment data
• Hardware reliability and maintenance costs
• Supply chain dependence for actuators and sensors
• Labour and regulatory pushback

Future outlook

Physical AI is the largest opportunity and the slowest to realise. Expect industrial and logistics use cases to scale first, with humanoids following where dexterity and mobility justify the cost.

Sources & references

1. 01Robot learning and vision-language-action model literature — Various (see Sources page)research
2. 02Robot learning and vision-language-action model literature — Various (see Sources page)research