Infrastructure against physical neglect.
Litter intelligence, live today · cleanup robotics, in build
Public space gets neglected in the gap between full sanitation crews and doing nothing — corridor edges, plaza spillover, park hotspots, drain approaches. Places too small or too frequent to clean often enough at conventional cost.
Two fronts. Trash Trail, live now, is the only public map that connects street litter to the drinking water it threatens — fusing resident reports, city 311 records, and EPA water-quality assessments with cited provenance. CrustBot, a route-class pickup robot in bench build, is where that data eventually points.
Operator-supervised. Cheap enough to be modular. Built to learn from real routes. We are early. Standards are operational, not aspirational.
Litter, mapped to the water it threatens.
Anyone can open the map and see litter reports over their watershed — which stretches the EPA lists as impaired for trash, and which drinking-water intake each spot drains toward. Signed in, you can report litter with photos and reachability tags, and log what you picked up on the cleanup ledger. Every number is cited.
Report
Photo + reachability tags + hazard pins. New reports reach other viewers' maps in seconds.
Log cleanups
"How much did you pick up?" — bag-bucket logging, crews & orgs, and a funder-shaped export for cleanup organizations.
Open data
The observation corpus publishes as versioned CC-BY-SA-4.0 GeoJSON snapshots with per-watershed source citations.
The robot is one of three layers. The data layer shipped first.
- 01 · live
Trash Trail
The coordination layer's public face: a litter map fused with city 311 records and EPA water-quality assessments, a cleanup ledger, and an open CC-BY-SA dataset. Every report is a data point a robot can eventually route against.
- 02 · bench build
CrustBot
The robot itself. Route-class, low and wide, lobster-style front intake, one dedicated dirty-work actuator package, Pi 5 + Pico 2 split. The design rule is dirty-path isolation — waste touches sacrificial parts, not the protected core.
- 03 · designed
Service operation
Routes, operators, site classes, battery swap and bag service. Before/after proof on every job is part of the operation, not an afterthought.
CrustBot — bench build.
A route-class machine with a lobster-style front intake and one cheap, dirty-tolerant actuator package. Compute is delivered and benchmarked; the frame build starts when the aluminum re-buy lands. No route, no pickups, no premature claims — the benchmark page is the scoreboard.
- Phase 0 bench build active
- Phase 1 first build compute delivered & benched · frame stock re-buy pending · one robot first
- Phase 2 first route no date
Frequency. Dignity. Proof. Serviceability.
Frequency
The failure mode isn't "never cleaned." It's "not cleaned often enough at the price point it would cost to do conventionally."
Dignity
Neglected physical space communicates neglect back to the people living around it.
Proof
Customers need evidence, not vibes. Every pickup is paired before / after. Every benchmark is reproducible.
Serviceability
Waste robotics only becomes real if the dirty parts are survivable and replaceable.
Where we are — August 2026
| Trash Trail | v1 live · 550+ observations · 16 DC/MD/VA watersheds tracked · open CC-BY-SA-4.0 commons |
| iOS app | Capacitor shell builds & runs in simulator · ready for App Store submission · enrollment pending |
| Hardware | Raspberry Pi 5 (1 GB) delivered & benched · Pico 2 (RP2350) safety MCU |
| Phase 1 | Drive & power staged · frame stock re-buy pending (supplier canceled) · one robot first |
| Perception | Pi 5 edge feasibility: YOLOv8n COCO baseline, 2.13 FPS sustained (2026-05-13) · litter detector v2 trained, serves the Trail photo pipeline |
| Simulation | Three.js + Rapier physics, firmware-in-the-loop · live viewer |
| Field pickups | 0 no chassis on a route yet |
What we're working on.
Trash Trail on the App Store
The iOS shell builds and runs; icon, screenshots, and in-app account deletion are done. What's left is Apple developer enrollment and review.
Pace estimates on every pile
The cleanup ledger logs how much got picked up; with enough logged history a site shows growth-vs-cleanup pace ("clear by October"). The estimator says "not enough data yet" rather than inventing a rate.
CrustBot frame build
The blocker on starting the frame build is frame aluminum — the supplier canceled the order. Re-buy, then cut and drill: the cutting tooling is on the bench already.
DregsBane Systems LLC is a Virginia company, formed April 2026, building infrastructure against physical neglect: Trash Trail, a public litter-and-water map that is live today, and CrustBot, a route-class cleanup robot in bench build.
It is deliberately small — everything on this site was designed, built, benchmarked, and deployed in-house. The build log is the honest record of who we are: what shipped, what didn't, and what the numbers actually say.
Contact: hello@dregsbane.com