New Run

Configure a problem instance and let the trained model plan a live, click-to-route orienteering path with time windows.

Configure a new route optimization

Generates a fresh problem instance (node locations, rewards, and time windows) and starts an interactive routing session driven by the trained Pointer Network.

Number of nodes

Locations to potentially visit, including the depot.

Time budget (hours)

Total time available before you must be back at the depot.

Beam width

How many candidate routes the recommender tracks at once.

Minimum reward decay

Fraction of a node's reward still paid if visited right at the deadline (0-1).

Cyclone-style forecast events

Rewards and evacuation deadlines shift on their own clock as a simulated hazard track updates, independent of your movement.

Random seed (optional)

Leave blank for a random instance each time.