Agent Frameworks

VeloCity's decentralized traffic AI cuts travel times across 4 global cities

Autonomous vehicles self-optimize routes via space-time reservations, beating gridlock in Tokyo and Manhattan.

Deep Dive

VeloCity is a decentralized multi-agent framework introduced in a new paper for connected and autonomous vehicles (CAVs) in arbitrary urban areas. Instead of centralized control, each vehicle queries a local traffic coordinator for a reservation table, independently computes its fastest conflict-free mobility profile, and reserves its space-time slots. Tested on four large-scale real-world urban maps—Tokyo, Manhattan, Rome, and Bologna—the framework drastically lowers travel times, tightly bounds delay variance, and prevents congestion gridlocks even under extremely high vehicular densities, all without scenario-specific tuning.

Key Points
  • VeloCity distributes route optimization to each CAV, using localized traffic coordinators for space-time slot reservations—eliminating global communication overhead.
  • Simulations on four real-world urban maps (Tokyo, Manhattan, Rome, Bologna) show drastically lower travel times and tightly bounded delay variance vs. state-of-the-art.
  • The framework requires no scenario-specific tuning and guarantees collision-free, physically executable trajectories even at extremely high vehicle densities.

Why It Matters

VeloCity could unlock scalable, congestion-free autonomous fleets in dense cities, slashing commute times for millions.

📬 Get the top 10 AI stories daily