Datasets:

alirezaaminzadeh
/

stochastic-last-mile-benchmark

benchmark_count int64	metrics dict	per_strategy dict
9	{ "best_cost_strategy": "ortools", "best_stability_strategy": "reinforcement_learning", "policy_accuracy_proxy": 0 }	{ "routefinder": { "avg_objective": 470646.439, "avg_stability": 0.838, "avg_distance": 6771.017, "avg_lateness": 57984.428, "avg_route_change_penalty": 0.536, "win_rate": 0 }, "rolling_horizon": { "avg_objective": 479628.636, "avg_stability": 0.83, "avg_distance": 7077.861, ...

Stochastic Last-Mile Re-optimization Benchmark

Benchmark results for adaptive last-mile routing under uncertainty, aligned with MBZUAI/svrp-bench.

File	Description
`manifest.json`	Dataset metadata
`eval_results.json`	Aggregate metrics per strategy
`lastmile_*.json`	Per-instance strategy comparison results

Scenario

100 delivery orders, 10 vehicles (medium preset)
Morning plan from RouteFinder (ai4co/routefinder)
Intraday disruptions: traffic, unavailability, new orders, vehicle failure
Re-optimization every 5 minutes or on disruption

Strategies Evaluated

RouteFinder, Rolling Horizon, ALNS, Monte Carlo, Scenario Tree, OR-Tools, Reinforcement Learning

Metrics

Metric	Description
`avg_objective`	distance + lateness + route_change_penalty
`avg_stability`	Schedule stability score (higher = less churn)
`avg_route_change_penalty`	Penalty for deviating from previous routes
`win_rate`	Fraction of instances where strategy wins

Objective

objective = distance + 8 × lateness + 12 × route_change_penalty

Related Resources

Citation

@dataset{stochastic_lastmile_benchmark,
  title={Stochastic Last-Mile Re-optimization Benchmark},
  author={LastMileAdapt Engineering},
  year={2026},
  publisher={Hugging Face},
  howpublished={\\url{https://huggingface.co/datasets/alirezaaminzadeh/stochastic-last-mile-benchmark}}
}