What it is
Edge computing runs computation on or near the devices generating data, such as a roadside gateway, building controller, or transit vehicle. It can filter, aggregate, or interpret data locally before sending selected results to a central platform. The design should state which functions continue during a network outage and which require remote services.
How the system fits together
Edge device
A gateway or computer collects nearby feeds and runs selected software within power, thermal, and maintenance limits.
Local data handling
Filtering, aggregation, or inference can reduce transmission, with retention and deletion behavior specified.
Fleet management
Operators track software versions, device health, security updates, and rollback across distributed units.
Cloud or control integration
Only necessary outputs travel onward, with clear authority for actions and behavior during connectivity loss.
- 01
Edge device
- 02
Local data handling
- 03
Fleet management
- 04
Cloud or control integration
Where cities use it
- Low-latency traffic and transit monitoring
- Local building controls
- Privacy-conscious video analytics
- Resilient utility or environmental sensing
Data generated
- Locally processed measurements
- Device and software health
- Aggregated counts or event alerts
- Audit and update records
Analytics and AI applications
- Computer vision
- Anomaly detection
- Real-time classification
- Bandwidth-aware sensor fusion
Example in practice
A roadside camera gateway computes vehicle and pedestrian counts locally and forwards only aggregated totals. The city documents the counting error, disables image retention, and tests whether counts remain available during network loss.
Design considerations
Choose sensor placement and refresh rates from an operational need. Keep device identity, calibration, network health, and location visible to the team using the data.
Limits and risks
- Distributed computers increase the number of devices to patch and secure.
- Local models may become inconsistent if versions and configurations drift.
- Faster response is not automatically safer or more accurate; edge decisions still need validation.