🚨 New approach to multi-agent reinforcement learning
[6.4.26] Researchers merged Multi-agent RL with Model Predictive Control so each agent learns strategy, but a model based controller still generates the final feasible action
The approach is slower to train, but stronger
🚨 Airdreamer - Generalist Drone Navigation with World Models
[6.2.26] Following recent advancements with UAV world model policies (like Skydreamer - vision based drone racing), researchers train a generalist policy for drone navigation
The world model works by learning to
Sam3D body was all over our timeline
So we decided to make our $100 FPV drone gesture controlled
> T-pose = takeoff
> arm right = go right
> arm left = go left
> arms up = track me
> arms crossed = land
FLIP is a new swarm-planning method that keeps 100 drones coordinating in real time by treating formation changes like point-cloud registration
In the paper’s benchmark, FLIP planned for 100 drones in ~0.040s mean time, compared with ~0.933s for Zhou’s method and ~5.814s for
New community hub just dropped for drone x ai enthusiasts
We'll be posting weekly:
> Changelogs
> SOTA research of the week
> Open sourced GitHub repos examples
Check it out:
thedroneforge.com/community
Welcome to Droneforge Email Newsletter
We’re going to start covering:
> weekly app changelogs
> top drone research of the week
> tutorials and guides for Nimbus
Want to stay updated? DM us your email and we’ll add you to the list