Edge AI

Edge AI
AI inference or learning performed near data sources

Edge AI refers to AI computation carried out on or near the devices that generate or use data, rather than relying exclusively on distant centralized computing. Edge processing can take place within a robot, sensor node, local gateway, industrial computer, or nearby network infrastructure.

Deployment models

At one end, an edge device runs a model trained in a data center. At the other, devices may adapt or train models locally. NIST describes different edge-AI arrangements depending on the role of edge nodes in inference and learning.

Benefits and trade-offs

Local operation can reduce round-trip communication and dependence on network availability for selected tasks. However, devices may have strict limitations on memory, energy, heat dissipation, and compute capacity. Updating models across many devices also introduces management concerns.

Use in physical AI

Robots often need prompt sensor processing and control decisions. Some tasks can be run locally, while heavier analytics and fleet-wide improvements can be handled elsewhere. This split is an architectural choice, not a guarantee of privacy or reliability.

Further reading

• AI Risk Management Framework — NIST

See also

• Physical AI
• Computer Vision
• Sensor Fusion
• Industrial Automation
• Autonomous Systems
• Machine Learning