Files
OKF_Vault/knowledge/challenges/prezero-waste-routing.md
T
Noel Sabia 7b9893a89e feat: add OKF knowledge bundle for hackathon challenges
Capture all six Future City Hackathon 2026 challenges as OKF v0.2 concepts, with dataset and organizer concepts, indexes, and a log.

Configure release-please with the simple strategy so the first release PR bumps the repository from 0.0.0 to 0.1.0.
2026-11-13 18:23:00 +01:00

6.9 KiB

type, title, description, resource, tags, status, generated, sources
type title description resource tags status generated sources
Hackathon Challenge WasteRouting.AI: fill-level-driven collection routes Analyze more than a year of glass-container fill levels, find the patterns behind them, and build a configurable route planner that recommends optimized collection runs. ../../PreZero/20261113_Readme%20Challenge%20PZ.docx
hackathon
future-city-hackathon-2026
prezero
waste-management
routing
optimization
time-series
logistics
stable
by at
opencode/deepseek-v4.1-flash 2026-11-13T17:13:56Z
id resource title author
prezero-readme ../../PreZero/20261113_Readme%20Challenge%20PZ.docx WasteRouting.AI — PreZero challenge README team:prezero
id resource title author
prezero-deck ../../PreZero/PreZero.pptx WasteRouting.AI — challenge pitch team:prezero

At a glance

Field Value
Organizer PreZero
Data More than a year of fill-level readings for glass recycling containers
Core problem Fixed area-by-area schedules ignore how full each container actually is
Output Data ingestion, pattern visualisation, and an optimized route engine
Key idea Make the routing model configurable through general operational variables
Scope A strong prototype concept is enough; refinement is expected after the event

The problem

The City of Heilbronn, like many municipalities, empties public waste bins area by area on a fixed schedule, regardless of how full each container actually is. In many cases fill-level data is neither collected nor tracked. Municipal services and private waste companies want AI route optimization, but those algorithms cannot deliver true efficiency without real-time fill-level intelligence.1

PreZero's in-house solution Fillinq can collect fill levels in a pragmatic, user-friendly way. The challenge starts where that data becomes available: what do the fill levels mean, and how should they change collection routes?1

Your task

1 — Analyze

Explore the fill-level dataset and uncover key insights:1

  • Patterns and cycles: What distinct accumulation patterns or cycles emerge over time?
  • Container grouping: Do specific containers share similar behaviour or fill rates?
  • Anomalies and external factors: How do school vacations, public holidays such as Christmas, Easter and Pentecost, and seasonal shifts affect waste volume?
  • Drivers: Look beyond surface trends to understand what causes the observed patterns.

2 — Develop

Turn the insights into actionable intelligence by engineering a tool that serves as the foundation for dynamic, optimized route planning.1

3 — Define general variables

Make the solution adaptable and scalable across municipal services by exposing configurable operational variables. The brief names:1

Variable Why it matters
Waste stream and density Different waste categories have different weight characteristics.
Container specifications Container type and maximum volume capacity constrain how much can be collected.
Fleet characteristics Vehicle types, volume capacities, and weight limits determine what a fleet can carry.
Inter-point distances Travel distance between individual collection points drives route cost.
Disposal depot proximity Distance from collection points to processing or unloading sites affects the route.
Depot distances Distance from the central yard to the first and last stops shapes the working day.
Shift constraints Maximum allowable working hours per driver or shift bound the route.

The list ends with an ellipsis in the brief; teams are expected to add variables they find relevant.

4 — Route optimization

Using these inputs, the system should automatically recommend optimized collection routes. Based on calculated total weight and volume, it should also determine the optimal fleet deployment, for example whether one vehicle or multiple vehicles are required.1

What you get

The package contains PreZero glass fill levels: 11,701 readings for 449 anonymized containers, covering 2025-10-17 to 2026-09-30, with three glass colours (white, green, brown).

Gaps to design around

The dataset is deliberately narrow. It contains only Location_id, material_type, date, and fill_level_perc.2

  • No coordinates. Container locations are anonymized IDs without latitude or longitude. Inter-point distances, depot distances, and route geometry are not in the file. A route engine needs an explicit assumption, a synthetic layout, or an external geodata source.
  • No container specifications. Capacity and container type are not provided; they are variables the brief asks you to model.
  • No fleet or depot data. Vehicle capacities, weight limits, shift rules, and depot locations must be configured or assumed.
  • Irregular readings. Fill levels are recorded at irregular intervals, not as a complete daily time series. fill_level_perc is mostly numeric from 0 to 100 but also contains the string overfull (1,192 rows), which needs an explicit encoding.
  • Anonymized locations. The IDs are stable within the dataset, so grouping and per-container histories still work, but they cannot be joined to a map without additional information.

Expected deliverables

The brief lists four deliverables:1

  1. Data ingestion tool: a functional interface to upload and process fill-level datasets.
  2. Pattern visualisation: dashboards illustrating trends, anomalies, and insights.
  3. Route engine: a working prototype that generates optimized route proposals.
  4. Documentation: a brief user guide covering the technical architecture and approach.

What makes a strong solution

The brief is explicit that the main goal is to see creative problem-solving and a prototype concept, not a finished product. Strong concepts could become the foundation for future development or even a startup project.1

A strong entry is therefore likely to:

  • Show real insight from the data rather than only a dashboard shell.
  • Make every operational assumption configurable instead of hard-coding one city's fleet.
  • Handle the missing spatial and fleet data transparently.
  • Connect pattern findings to routing decisions, not treat analysis and optimization as unrelated demos.
  • Explain why the proposed routes are better than a fixed area-by-area schedule.

Sources

The README is the detailed brief.1 The pitch deck states the same challenge in short form.3 Dataset facts come from the delivered CSV.2


  1. PreZero, "The Challenge — WasteRouting.AI", challenge README, 2026. ↩︎

  2. PreZero, glass_fill_levels_anonymized_f.csv, delivered dataset, 2026. ↩︎

  3. PreZero, "WasteRouting.AI", challenge pitch deck, 2026. ↩︎