Swarm Drone Urban Pathfinding Algorithm

I need an implementable approach that lets a swarm of autonomous drones discover the most efficient path through AI technologies in a busy urban landscape while constantly negotiating dynamic obstacles such as cars and pedestrians. The emphasis is on efficiency rather than simple collision-free travel, so your solution should balance travel time, energy usage and real-time rerouting without spreading the drones too far apart.

I am open to a bio-inspired technique—ant colony optimisation, particle swarm, artificial bee or a hybrid—so long as it scales cleanly from small (5–10 drones) to medium fleets (50+). Please choose libraries and simulation tools you are confident with; ROS 2, PX4, Gazebo, Webots, Python fine if they help you deliver quickly.

Deliverables
• Well-documented source code of the swarm path-planning algorithm
• A repeatable simulation showing successful urban navigation with moving vehicles and pedestrians randomly injected into the map
• Short report (2-3 pages) explaining design decisions, parameter tuning and quantitative results (average flight time, energy consumed, success rate)

Acceptance criteria
• 95 % or better mission completion rate over 100 simulation runs
• Average flight time no more than 10 % longer than the single-drone theoretical optimum on the same map
• No mid-air collisions or deadlocks under maximum load conditions

If you have previous work on multi-agent path finding or reinforcement learning for robotics, please mention it; otherwise, a concise proposal outlining your chosen technique and timeline is enough to get us started.