.png&w=256&q=75)
Adam Ta@Sifr
Sifr
🤓 Latest Submissions
DroneMD
DroneMD is a GPU-accelerated drone swarm simulation platform for designing, validating, and analyzing large-scale drone formations. Starting from any 3D model, DroneMD generates a swarm formation, simulates physically accurate drones, visualizes the swarm in real time, and produces an interactive safety report before deployment. By providing realistic simulation and visual data, DroneMD enables reinforcement learning research and autonomous navigation development without requiring costly or potentially unsafe real-world drone flights. Workflow 1. Import a 3D Formation - Upload a custom OBJ model or select a built-in formation. - Automatically generate drone target positions from the geometry. 2. Configure the Simulation - Adjust swarm behavior using physical parameters (separation, alignment, cohesion, goal seeking, and obstacle avoidance). - Configure boundary conditions, physics fidelity, integrator, and swarm size. 3. Run GPU-Accelerated Simulation - Simulate each drone as a physically accurate Crazyflie quadrotor with rigid-body dynamics. - Physics are executed on the GPU using JAX, and CrazyFlow enabling real-time simulation of large swarms. 4. Visualize the Swarm - Monitor the simulation through an interactive Three.js interface. - Inspect drone trajectories, collisions, agent states, and formation evolution from any viewpoint. - Capture realistic visual observations that can be used to develop and evaluate reinforcement learning policies and perception algorithms entirely in simulation. 5. Analyze Safety and Performance - Receive an interactive report summarizing the simulation. - Evaluate safety score, collision rate, formation accuracy, swarm connectivity, average speed, kinetic energy, and other performance metrics before real-world deployment. 6. Export Results - Export trajectories and reports in CSV, JSON, ROS waypoints, or PDF formats for further analysis, reinforcement learning pipelines, or deployment.
13 Jul 2026
