Aerbits: A Timeline and the State of Things

People ask me what Aerbits is, where it came from, and how it's going. The honest answer doesn't fit in an elevator pitch, so I'm writing it down. This is the whole arc: how a frustration on my own street in San Francisco turned into a drone and AI platform, where it stumbled, and how it just became, this spring, the first system of its kind a city in America has voted to put to work.

Why Aerbits exists

It started with trash. Around 2021 I was walking my southeastern San Francisco neighborhood with my kids, around and over piles of illegal dumping that nobody seemed to be dealing with. The city does clean it up when it's reported. But reporting is sporadic and incomplete, so countless dump sites simply sit there, unseen and unaddressed. The system wasn't broken because nobody cared. It was broken because nobody had consistent, accurate visibility.

The premise of Aerbits has never changed: use drones and computer vision to find illegal dumping in real time, persistently and accurately, and put that information directly in front of the people who can act on it. As Oakland Councilmember Zac Unger put it when the city signed on, "Success wouldn't be how much trash we collect. Success would be how much trash we don't see."

A brief but complete timeline

2021: The seed. The idea crystallizes out of personal frustration with uncollected waste in my neighborhood.
2022: I jump. After nearly a decade at Trulia and Zillow, I leave to start Aerbits full time. First drones in the air, first detection models, first end-to-end pipeline.
May 2023: Going public with it. I present "Daily Detection of Illegal Dumping" at IDCon 2023, held, fittingly, in Oakland.
2022 to 2023: The Bayview proof. A self-funded pilot in San Francisco's Bayview-Hunters Point cut active dumpsites by 96%, from 118 down to 5, over a 26-day window. A broader three-month, seven-neighborhood case study filed roughly 1,500 reports (525 large piles, 264 bagged piles, 98 furniture piles, 26 mattresses, 21 toxic spills, and more). The concept worked. San Francisco didn't continue it.
November 2, 2023: Patent published. The USPTO publishes the application for identifying trash within a geographic boundary using unmanned aerial vehicles.
December 2023: The architecture and the vision. I document the full system and write down the ten-year view of where this all leads.
January 2026: The rebuild. With three years of operational learning and a computer-vision field that had transformed underneath me, I started rebuilding Aerbits from the ground up.
January 2026: Oakland comes calling. Oakland reaches out about its illegal-dumping problem.
February 2026: Building the proposal. We start putting together a formal proposal for the city.
March to April 2026: Through the process. A Privacy Advisory Commission review (March 5), a unanimous Public Works & Transportation Committee vote (March 24), and a unanimous City Council approval of the pilot (April 14), capped by Mayor Barbara Lee announcing the program at an April 20 press conference. Oakland becomes the first city in the nation to formally back Aerbits.
Today (June 2026): In active procurement. We're working through procurement with Oakland Public Works to finalize the contract and get drones in the air.

What Aerbits actually does

The shape of the system has been consistent, even as the pieces underneath have been swapped out for far better ones.

Capture. Enterprise drones (DJI Matrice 4E, fitted with an AVSS parachute safety system) with high-resolution cameras and RTK GPS fly planned routes, shooting overlapping imagery accurate to within centimeters. One drone covers roughly a square mile every 30 minutes.
Detect. A proprietary computer-vision model, trained on our own dataset of 100,000+ drone images, finds waste in every frame: mattresses, furniture, tires, appliances, construction debris.
Locate. Each detection is projected from pixel space onto real-world latitude and longitude using the drone's GPS, orientation, focal length, and ground elevation, so a pile in a photo becomes a precise dot on a map, classified by severity and waste type.
Report. Reviewed detections flow out through a two-way 311 integration, filing accurate, well-located, image-backed service requests with GPS coordinates and timestamped aerial photos attached.

That's the core loop: capture, detect, locate, report. Everything in the rebuild is about making each step sharper and adding what comes after the report.

Built for the public's trust

A camera in the sky over a city only works if people trust it, so Aerbits is deliberately not a surveillance tool. There is no facial recognition and no license-plate reading. It photographs public rights-of-way only. In Oakland, unredacted images are deleted within a week (redacted ones kept up to six months), there's an Oakland-exclusive clause so the imagery is never reused for other clients or model training, and the program is subject to quarterly audits and annual public reporting. The goal is finding trash, not watching people.

The rebuild, and where the tech stands

The original Aerbits proved the concept. The January 2026 rebuild is about turning a proof into a platform.

A new backend supports multiple datasets and detection classes, real-time pipelines, and a flexible model layer tuned per task. Pixel-level segmentation moved us from bounding boxes to real understanding: tracing the exact boundary of a pile lets us estimate area and volume, separate adjacent items, and track how a site changes over time. The dataset refinement project, hand-correcting the entire historical dataset with three years of hard-won intuition, is the least glamorous and most important work happening right now. Semantic masking teaches the system what to ignore (private property, active job sites) as carefully as what to find. And the loop now closes all the way to the field: fleet management plus a capacity-aware route optimizer turn a map of detections into efficient pickup routes, while a vision-language layer handles the judgment calls around detection, like whether something is a construction site or dumping, whether a pile has grown, and whether we can draft a clean 311 report for a human to approve in seconds.

The patent

The patent application, "System and Method for Identifying Trash Within a Predetermined Geographic Boundary Using Unmanned Aerial Vehicles," was published in November 2023 and covers the drone-based detection approach at the heart of Aerbits. It frames the problem along four lines of attack: reactive (find dumping consistently and quickly), predictive (anticipate where it will appear), preventative (be present where and when it's likely, to deter), and punitive (use surveillance to identify and prosecute repeat offenders).

So how's it going?

For the first time, the honest answer is: really well, and in public.

This spring, Oakland became the first city in the country to formally back Aerbits. Oakland reached out in January, we spent February building a proposal, and the next two months went through the city's process: a Privacy Advisory Commission review, the Public Works & Transportation Committee, and a unanimous City Council vote on April 14 to fund a $150,000 six-month pilot (72 flights across roughly 1,440 linear road miles, paid out of the Comprehensive Clean-Up Fund with no impact to the general budget). Mayor Barbara Lee announced it at an April press conference. We're now in active procurement with Oakland Public Works to finalize the contract and get drones in the air. It's part of a broader crackdown the city passed at the same time, which raised illegal-dumping fines to $1,500, $3,000, and $5,000 for repeat offenses. The need is enormous: Oakland fields more than 25,000 dumping reports a year, about 70 a day, and a complaint-driven system can't keep up. Aerbits makes it data-driven instead.

That breakthrough makes the earlier disappointment worth it. The Bayview pilot showed the technology works, and a 96% reduction in active dumpsites is not a rounding error, but San Francisco didn't continue the program, and for a while it wasn't clear anyone would. I kept flying, kept collecting, kept building, and rebuilt the whole platform on the bet that the problem and the technology would eventually meet a city ready to act. Oakland was that city.

There's a long way to go. A pilot is a beginning, not a finish line, and the real test is the numbers Oakland sees on its own streets over the next six months. But the streets that started all of this are cleaner than they were, the technology finally matches the ambition, a city has voted to put it to work, and I'm just getting started.