01 / overview

Every road you manage, watched the moment you say so

Roads Management Insights (RMI) turns the roads you already manage into a live, historical, and safety-graded picture of what's actually happening on them — real travel time, disruptions as they form, actual traffic volume, even near-miss braking events — without a single sensor of your own. You define the routes; Google's Maps traffic network and Google Research's Mobility AI do the watching.

See how routes get captured How it's delivered

in one line

What it unlocks: pick any road segment you manage, and get its congestion, disruptions, volume, and safety signal on tap — no field crew required.
Where it comes from: Google Maps' live traffic network, fused with Google Research's hard-braking crash-risk signal.

RMI overview docs

your_project.rmiper selected route

historical_travel_timeTABLE

recent_roads_dataTABLE

disruptionsPREVIEW

vehicle_countsPREVIEW

···· hard-braking events, delivered alongside

0real-time refresh (Pub/Sub)

0bulk refresh (BigQuery)

0of history retained in BigQuery

0jurisdiction buffer for edge routes

02 / capture your own routes

You choose the roads — RMI does the watching

Nothing gets tracked until you say so. Submit an origin, a destination, and any waypoints as coordinates through the Roads Selection API, and RMI validates the route against two gates before it will report anything on it: is this road inside your approved jurisdiction, and does it carry enough real traffic to report on without exposing individual drivers.

the two gates

Jurisdiction: one or more polygons defining your approved territory — routes can extend up to ~20 km beyond it with verification, but not much further.
Road utility: a minimum-traffic-volume threshold, so quiet residential streets never get published — this is a privacy floor, not a bug.

Coverage isn't a one-time approval: RMI re-checks traffic volume every quarter, and a route that stays quiet for four consecutive quarters is automatically retired.

Usage guidelines & limitations

Origin, destination, waypointssubmitted as lat/lng via the Roads Selection API

Jurisdiction checkis this road inside territory you manage?

Road utility checkdoes it carry enough traffic to publish safely?

Active tracked routedata collection begins from this point forward

03 / traffic & travel time

Real travel time, and how much worse than free-flow it really is

Two signals underpin everything else RMI ships. Trip duration comes in two flavors per segment — traffic-aware and static (free-flow) — so a straightforward ratio tells you exactly how congested a road is right now versus its baseline. Speed Reading Intervals (SRI) layer a simple traffic-density read on top: normal flow, slowdown, or severe delay. Click run below to see the pattern.

congestion ratio

rank your routes by how much traffic is slowing them down

SELECT
  route_id,
  route_name,
  duration_in_seconds,
  static_duration_in_seconds,
  ROUND(duration_in_seconds / static_duration_in_seconds, 2) AS congestion_ratio
FROM rmi.historical_travel_time
WHERE observed_at >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 1 HOUR)
ORDER BY congestion_ratio DESC
LIMIT 5

Query complete · 3.1s elapsed · 5 rows

route	ratio	read
7th Ave / Downtown Connector	2.4×	severe delay
Route 9 / River Crossing	1.9×	slowdown
Airport Corridor NB	1.5×	slowdown
Industrial Loop East	1.1×	normal flow

speed reading intervals

same window, categorized by SRI band

SELECT
  sri_band,
  COUNT(DISTINCT route_id) AS routes_affected
FROM rmi.recent_roads_data
WHERE updated_at >= TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 15 MINUTE)
GROUP BY sri_band
ORDER BY routes_affected DESC

Query complete · 1.8s elapsed · 3 rows

SRI band	routes affected
Normal flow	142
Slowdown	31
Severe delay	6

Real-time road data docs

04 / disruptions

Know about the crash before the backup reaches you

RMI's disruptions layer identifies road closures directly, and goes further with Early Signal Detection — surfacing likely crashes and stalled vehicles from traffic pattern anomalies, often before a mile of stopped traffic shows up on anyone's dashboard. It's built for exactly one job: giving a dispatcher or planner the extra minutes that matter.

what it enables

Closures identified directly, layered onto your tracked routes.
Early Signal Detection infers probable incidents from how traffic is actually behaving, ahead of confirmed reports.
Safety-aware dispatching — roadside assistance and utility crews can be routed around high-risk zones and secondary-accident queues.

Disruptions & Early Signal Detection announcement

Traffic pattern anomalyspeed drops faster or harder than normal congestion

Early Signal Detectionflags a likely crash or stalled vehicle

Disruption publishedappears in your route's live disruption feed

Safety-aware reroutedispatch avoids the zone before it's confirmed

05 / vehicle counts

Turn "we think it's busy" into a defensible number

The vehicle counts dataset gives you actual traffic volume for the routes you track — the number capital projects, grant applications, and safety studies actually require, instead of a manual count from a study conducted years ago.

Justify capital projectsBack a widening, signal, or repaving budget with current, objective volume data.

Prioritize safety fixesCorrelate volume against historical incident data to find where a fix has the most impact.

Strengthen grant applicationsObjective volume figures replace estimates that reviewers routinely discount.

Track change over timeSee whether an intervention actually shifted traffic volume, not just travel time.

Vehicle counts announcement

who asks for this number

Capital planning teams Grant reviewers Road safety auditors Public works budgeting

replaces

Manual tube counts Stale traffic studies Rough estimates

06 / hard-braking events

A near-miss happened here — long before it's a police report

Built by Google Research's Mobility AI team, hard-braking events (HBE) are aggregated, anonymized signals of sudden, aggressive deceleration — a leading indicator of crash risk that's far more frequent, fresher, and denser than lagging collision statistics. Congestion, lane merges, toll plazas, and complex interchanges reliably produce harsh braking, and harsh braking reliably precedes crashes.

why it beats crash reports alone

near-misses happen far more often than crashes, so the dataset is denser and updates faster
lets you find high-risk segments before they've generated enough official reports to show up as a hotspot
externally validated: research on New Jersey interstates matched 8.5M+ connected-vehicle telemetry records against ~45,000 police-reported crashes

Google Research: hard-braking events blog

Sudden deceleration Aggregated & anonymizedno individual driver or trip identified Hard-braking event signal

HBE density by segment Leading indicatorpresent weeks or months before a crash cluster High-risk segment, flagged early

07 / delivery & architecture

Two speeds: build the case, or run the operation

RMI ships through two packages that match two different jobs, both riding on the same underlying schema — so a query built for one translates directly to the other. On top of that, an ADK agent lets planners ask questions in plain language instead of writing SQL by hand.

Foundations

Delivered via BigQuery Analytics Hub
Refreshes every 10 minutes
Built for historical analysis, planning, reporting
Up to 10 years retained before deletion

Real-Time Operations

Streamed via dedicated Cloud Pub/Sub topics
Refreshes roughly every 2 minutes
Built for live dispatch, dynamic rerouting
Protobuf / JSON payloads per route

BigQuery and Pub/Sub usage bill separately from the RMI license itself — standard BigQuery storage/query rates apply, and Pub/Sub subscriptions run about $50/TiB with 10 GB free per month.

ask it in plain language

the RMI ADK agent pairs a conversational root agent with a BigQuery sub-agent that writes and runs the SQL
runs on Gemini, querying the same historical_travel_time and recent_roads_data tables shown above

Real-time road data docs RMI ADK agent architecture

08 / who it's for

Built for the roads you already manage — now open to the fleets that drive them

RMI started with city and regional planners, road authorities, and public utility providers seeking to understand and respond to traffic conditions on their own networks. As part of Google Earth AI, it's now expanding to commercial fleet operators — connecting the agencies that manage roads with the private companies driving on them every day.

access today

public sector entities — road authorities, utilities, cities and regions — are eligible now
private-sector fleet access is expanding via Google Earth AI, with new datasets rolling out in preview

RMI overview docs RMI availability announcement

Safety triageFind high-risk segments from HBE density before crash reports pile up.

Capital planningJustify projects and grants with real vehicle-count and congestion data.

Live dispatchReroute crews around disruptions the moment Early Signal Detection flags them.

Fleet operationsPrivate carriers gain the same congestion and disruption picture the roads' own managers see.