Leaderboard
130 tasks · 13 city maps · 60 turns · 3 repeats
| Agent | Net incomeUSD / task ↑ | Deliveredorders / task ↑ | On time% ↑ | Red-light violationsevents / task ↓ | Obstacle collisionsevents / task ↓ |
|---|
Evaluation details
Waypoint selects visible registered destinations; Any-Point selects an image point that projects to a reachable location. Scores average three repeats per task, then weight tasks equally. On-time rate covers completed deliveries. Oracle is a simulator-assisted search reference.