Engine
GPU vehicle routing
cuOpt solves capacitated, time-windowed routing with pickups, deliveries and multiple depots.
Routite Optimize · GPU route solver
Optimize is the engine. It uses NVIDIA cuOpt on GPUs to re-solve routing, dispatch and scheduling with every constraint you run on, starting from the plan drivers are already following.
Inside a re-solve
Re-solve summary · Zone 4, 10:47
ExampleStop 01 · Capabilities
Engine
cuOpt solves capacitated, time-windowed routing with pickups, deliveries and multiple depots.
Stability
Each re-solve starts from the current plan and penalises needless changes.
Scope
Re-plan the full network, not one driver at a time, so stops can move between vans.
Objectives
Weight distance, lateness, overtime and emissions; change weights per region.
Explain
Each moved stop carries why it moved, ready for Copilot to explain.
Scenarios
Run the same day with one more van or a different cut-off and compare.
Stop 02 · Constraint catalogue
Filter by kind of rule and by release. Anything missing for your fleet is the first thing to raise in a pilot.
| Constraint | What it means | Kind | Release |
|---|---|---|---|
| Time windows | Hard or soft windows per stop, with lateness penalties. | Time | Core |
| Vehicle capacity | Weight, volume, pallets or any count, per compartment. | Load | Core |
| Driver hours | Shift length, breaks and maximum drive time. | Time | Core |
| Skills | Jobs that need a certified technician or a two-person crew. | People | Core |
| Vehicle type | Stops that need a tail-lift, a small van or a reefer. | Load | Core |
| Pickup and delivery | Paired stops in the right order on the same vehicle. | Load | Core |
| Multi-depot | Vehicles that start and end at different depots. | Network | Core |
| Zones | Keep drivers inside familiar areas unless it pays to cross. | Network | Planned |
| Priority orders | Orders that must be served even if others drop. | Time | Core |
| Temperature | Chilled and frozen compartments with door-open limits. | Load | Planned |
| Change penalty | Prefer plans that leave on-track drivers alone. | People | Core |
| Emissions | Weight CO₂ per km by vehicle and add it to the objective. | Network | Planned |
No constraints match both filters. Choose another kind or release.
Stop 03 · How it works
All orders, vehicles and constraints are solved before the shift.
Twin hands over the changed state; Optimize re-solves from the current plan.
Moved stops, new ETAs and reasons are passed to Copilot for review.
Stop 04 · Specification
| Item | Detail | Status |
|---|---|---|
| Solver | NVIDIA cuOpt | In build |
| Problem types | VRP with time windows, capacities, pickup and delivery, multi-depot | In build |
| Regional re-solve | About 1,000 stops in under 10 s | Design target |
| Full-day solve | Up to 10,000 stops per region | Design target |
| Objectives | Distance, lateness, overtime, emissions | In build |
| Deployment | Routite cloud; VPC and on-premises | Planned |
Stop 05 · Place in the loop
Optimize takes the live state from Twin and hands its plan to Copilot for review. It is the step every other product depends on.
Return leg Edge sends real arrival and service times back to Twin. ETA and demand models retrain on them, so tomorrow's first plan starts closer to reality.
Stop 06 · Questions
It can be. Many teams will start by running it beside the current planner during a pilot and comparing.
Tell us. Encoding unusual business rules is part of every pilot.
No. Routite cloud runs them. On-premises deployment on your own GPUs is planned.
Stop 07 · Destination
Next step
A replay of one past day shows what this product would have done, with every assumption written down.