Technology — HOTMESS

Architecture

A software layer above existing infrastructure

HOTMESS doesn't replace any existing systems. It sits above them as a coordination layer, reading sensor data and recommending network-wide optimisations.

Coordination layer

New capability

HOTMESS — Network-wide optimisation engine NEW

Control system

Existing, unchanged

SCATS — Adaptive signal control (intersection-level)

Sensors

Existing, already deployed

Inductive loops · CCTV cameras · Bluetooth detectors

Infrastructure

Physical assets

Traffic signals · Road network · Vehicles

Data inputs

Using sensors cities already own

HOTMESS fuses data from multiple sensor types - all of which are already deployed and maintained. None currently used for automated network analysis.

Inductive loop detectors

Already embedded at signalised intersections in most urban networks. Detect vehicle presence, count, and occupancy. The primary input for SCATS adaptive phase control.

CCTV cameras

Traffic cameras deployed across most urban networks are typically human-monitored only. HOTMESS applies computer vision for automated queue length estimation, turning counts, and vehicle classification.

Bluetooth & emerging sensors

Bluetooth detectors measure real origin-to-destination travel times by tracking anonymous device signatures - providing ground-truth network performance data. The architecture also supports integration with emerging sensor types such as LiDAR, radar, and thermal imaging as they become available.

Core approach

Reinforcement learning for network coordination

Traditional signal control optimises each intersection independently. HOTMESS treats the entire signal network as a single optimisation problem.

Using reinforcement learning, the system learns to coordinate signals across corridors and networks by observing real traffic patterns and discovering timing strategies that minimise total network delay - not just delay at individual approaches.

The model trains first on a digital twin built from historical SCATS data, then validates against real-world outcomes in shadow mode before any recommendations reach operations teams.

Observe Network state from loop, CCTV & Bluetooth sensors

↓

              Decide
              RL agent selects network-wide signal timing adjustments
            

↓

Act Recommendations surfaced to human operators

↓

Learn Measure outcomes, update model to reduce total network delay

↻

Safety architecture

Built on safety-critical engineering principles

The team's background spans aerospace, rail signalling, and major infrastructure - industries where safety assurance is non-negotiable.

Human in the loop

Operations teams retain full control. HOTMESS recommends, humans decide. No autonomous changes to signal operations without explicit authorisation.

Graduated trust

The system earns operational responsibility through evidence, not promises. Each phase requires demonstrable performance before progressing.

Transparent decisions

Every recommendation includes its reasoning and predicted outcome. Operators can see why, not just what. Full audit trail for every suggestion.

Fail-safe design

If HOTMESS is disconnected or fails, SCATS continues operating exactly as it does today. The coordination layer is additive, never load-bearing.

Measurable outcomes

What network-wide coordination can deliver

Network signal coordination directly impacts journey times, emissions, and safety. These outcomes are measurable from day one of a shadow-mode pilot.

Reduced journey times

Network-wide coordination reduces unnecessary stops and smooths traffic flow, cutting total travel time across corridors.

Lower emissions

Fewer stops and less idle time means less fuel burned and lower emissions. Transport is responsible for 85% of urban NOx.

Improved safety

Smoother flow reduces conflict points and aggressive driving behaviour associated with stop-start congestion patterns.

How HOTMESS works