Language / 语言: English primary · 中文概览如下。
本地优先的配送路线与派单参考引擎，提供带明确约束、可复现的优化逻辑。
A local-first reference engine for delivery routing and dispatch optimization.
CUSTOMER DECISION · APPLIED REFERENCE · BENCHMARK EVIDENCE
Decision question: How can orders be assigned and sequenced locally with explicit constraints, reproducible routing logic, and no mandatory paid dispatch API?
Delivery dispatch needs an executable optimization layer:
orders + workers + capacity + geography
↓
assignment
↓
route ordering
↓
route geometry / metrics
The project includes an Amap-compatible API shape, but "free alternative to Amap" should not be its identity. API compatibility is an integration feature; the core asset is the local-first dispatch engine.
GCJ-02 request
↓
coordinate normalization
↓
spatial clustering
↓
worker / cluster assignment
↓
route optimization
↓
OSM / routing distance layer
↓
Amap-compatible response + geometry
Current components include DBSCAN, OR-Tools, OSMnx / routing fallbacks, coordinate conversion, FastAPI, and a built-in map.
Current benchmark:
- 20 random datasets;
- 20 orders per dataset;
- Beijing Chaoyang, 5 km zone.
Reported results:
Metric Open Dispatch OSRM Trip Interpretation
Euclidean distance 19.0 km 19.9 km 4.8% shorter on of produced this benchmark ordering
The correct conclusion is:
The produced ordering was shorter under the chosen Euclidean comparison on this benchmark, while the engine's road-distance estimates were materially lower than OSRM and require calibration / stronger routing fidelity.
Do not summarize this as "we're better than OSRM."
- the local engine can produce competitive route orderings in the published small benchmark;
- Amap-compatible integration can be implemented without making a paid API the optimization core;
- local routing can support privacy / cost / deployment flexibility.
- better real road travel time than OSRM;
- better dispatch business outcomes than commercial dispatch systems;
- production-grade map / turn restriction fidelity;
- universal 4.8% advantage.
Use logistics-dispatch-clustering for learned operational priors and
this repository for explicit optimization.
historical behavior
↓
learned priors
+
explicit constraints
↓
Open Dispatch
↓
executable plan
Keep the existing FastAPI start command and Amap-compatible example, but
move the full request schema to docs/api.md.
- benchmark against OR-Tools routing with the same distance matrix;
- compare road-network travel time, not only Euclidean sequence length;
- real operational replay;
- capacity / SLA / overtime metrics;
- route stability across re-plans;
- calibrated distance / time error.
topprism:
purpose: customer-decision
capability: delivery-dispatch
platform_layer: decision-engine
maturity: applied-reference
evidence:
type: benchmark
scope: "20 random datasets × 20 orders, Beijing Chaoyang"
related:
- logistics-dispatch-clustering