A Breakthrough Worth Celebrating
In its Spring 2026 issue, Esri's ArcUser magazine published a feature story on how the Los Angeles Sanitation & Environment department (LASAN) is using AI to keep the city clean. The article, “AI Helps Clean Up LA,” by Jim Baumann, describes how LASAN upgraded its CleanStat program by mounting cameras on the city's ~900 trash collection trucks and feeding the footage into YOLO computer vision models.
This is excellent news — not just for LA, but for anyone who believes that technology can make cities cleaner. A leading GIS publication profiling an AI-powered detection system at municipal scale is a powerful signal that the approach we and others have been developing is moving from pilot phase to mainstream practice.
But what’s most interesting isn’t what the article describes. It’s what happens when you ask: what if you could do this from the ground and from the air?
What LASAN’s System Does Well
LASAN’s CleanStat 3.0 is an impressive deployment. Here’s what it achieves:
- Massive scale: Cameras on 900 trucks capture street-level imagery across LA’s 470 square miles
- Automated detection: YOLO models identify trash bins, tents, construction cones, and illegal dumping with confidence scores in the 0.79–0.96 range
- GIS integration: Detections flow through ArcGIS GeoEvent Server into operational dashboards with Google Street View verification
- Frequency leap: Weekly assessments instead of the previous 6–8 week manual cycle
- Field crew efficiency: Crews freed from surveying to focus on actual cleanup
“The YOLO model can correctly identify objects that are outside of LASAN’s interest, showing that this technology can be used for various departments.”
Truck-mounted cameras are a smart deployment strategy: they leverage existing city fleet infrastructure, they run on a predictable schedule, and they capture the street-level view that’s most relevant for sanitation crews. For cities that already have collection trucks running regular routes, this is a natural first step.
Where Truck Cameras Have Blind Spots
As good as the truck-mounted approach is, it has inherent limitations that come with operating at ground level:
- Line of sight: A camera mounted on a truck sees what’s visible from the street. Items behind parked cars, in recessed doorways, behind walls, or set back from the curb are invisible or partially obscured.
- Narrow field: The camera captures the width of the street and immediate sidewalk. Alleys, vacant lots, green strips, and utility corridors between buildings fall outside the detection zone.
- Volume measurement: A ground-level photo provides limited information about the three-dimensional volume of a dumpsite. Is it a bag of trash or a truckload?
- Waste composition: Identifying specific waste types (construction debris, hazardous materials, organic waste) from a single ground angle is challenging when items are piled or concealed.
These aren’t flaws in LASAN’s system — they’re inherent properties of ground-level sensing. Every sensing modality has blind spots. The question is how to cover them.
What Drones Add to the Picture
Aerial drone detection doesn’t replace truck-mounted cameras — it fills the gaps they can’t reach. Here’s how the two approaches compare across key dimensions:
- Street-level line of sight
- Follows truck route pattern
- 2D image from ground angle
- GPS from truck (road-level accuracy)
- Weekly pass on collection schedule
- Leverages existing fleet
- Bird’s-eye overhead perspective
- Flexible flight path (anywhere, on demand)
- 3D volume measurement from altitude
- RTK/PPK centimeter-level GPS accuracy
- Hourly/daily as mission requires
- Reaches areas no truck can access
The difference in location accuracy is particularly important for operational efficiency. A truck’s GPS places an object within a few meters of the vehicle’s position. Aerbits’ RTK/PPK-enabled drones achieve centimeter-level positioning, meaning cleanup crews arrive at the exact coordinates rather than searching a block for “somewhere near the corner.”
Behind Vehicles
Drones see items hidden behind parked cars, trucks, and street furniture that ground cameras miss entirely.
Full Sidewalk Coverage
From above, the entire sidewalk corridor is visible — including recessed areas, set-backs, and property edges.
Inaccessible Areas
Alleys, vacant lots, utility corridors, creek beds, and rail rights-of-way — places no truck route reaches.
Better Measurements, Better Decisions
One of the most significant advantages of aerial detection is the quality of the measurements it produces. A drone flying at altitude captures a dumpsite from multiple angles in a single pass, allowing the AI to measure not just presence but scale.
A ground-level camera might detect that a dumpsite exists. An aerial system can estimate its volume in cubic yards — the difference between a bag of household trash (a 10-minute pickup) and a construction debris pile that requires a dump truck and an hour of crew time. That distinction is the difference between efficient routing and wasted trips.
Similarly, multi-angle aerial imagery provides better data for waste composition classification. Construction debris reflects differently than organic waste. Hazardous materials have distinctive spectral signatures. The overhead perspective gives the AI model more visual information to work with, producing more reliable classification.
The Synergy: Ground + Air
The most powerful approach isn’t choosing one modality over the other. It’s deploying both.
Here’s how a combined system works in practice:
- Truck cameras provide weekly street-level sweeps across the entire city, catching the majority of curb-adjacent problems at low marginal cost per mile.
- Aerial drones provide targeted, high-resolution coverage of known hot spots, hard-to-reach areas, and follow-up verification — on whatever schedule the situation demands.
- Data fusion combines both feeds into a single operational dashboard. Crews see a unified picture: truck detections along their routes, drone detections in the back alleys and vacant lots, all with precise location data and photographic evidence.
A city using both approaches gets the breadth of ground-level fleet coverage plus the depth of aerial intelligence. Truck cameras keep a baseline watch on the entire street network. Drones zoom in on the places trucks can’t see, measure what ground cameras can’t measure, and verify cleanup completion with before-and-after imagery.
Real-World Precedent: Bayview-Hunters Point
Aerbits’ 13-month Bayview pilot demonstrated what consistent aerial monitoring can achieve: a 96% reduction in active dumpsites from a baseline of 118, verified through a controlled A-B-A withdrawal study. The system filed 4,376 automated 311 reports, and 30–50% of detected sites had never been reported through traditional channels.
Combined with a truck-based system, that detection gap narrows further — and the cleanup data becomes richer, more complete, and more operationally useful.
A Validation Moment for the Industry
LASAN’s CleanStat 3.0 deployment, as profiled in ArcUser, marks an important milestone. The nation’s second-largest city has operationalized AI-powered object detection for street cleanliness, using Esri’s enterprise GIS stack, and is publishing the results. This lowers the barrier for every other city considering similar technology.
But the most advanced deployments won’t stop at truck cameras. They’ll layer ground and aerial detection into a comprehensive system that sees more, measures better, and responds faster than either approach could alone.
At Aerbits, we’ve been building the aerial half of that stack for years — with the precision, coverage, and data quality that cities need to move from complaint-driven cleanup to intelligence-driven operations. The ground-level work happening in LA validates that the approach works at scale. The next step is putting both pieces together.
If LA’s future looks cleaner — as the ArcUser article concludes — imagine what a city with eyes at every level will look like.