Aerial Gym Simulator: A Framework for Highly Parallelized Simulation of Aerial Robots

ITH increasing deployment in a vast range of applications, including inspection, delivery, and search-and-rescue, aerial robots have gained immense popularity. Multi-rotor systems of varying scales have taken diverse roles and forms ranging from large vehicles with significant payload-carrying capacity to racing micro drones and reconfigurable robots capable of changing their shape actively or passively for traversal [1]-[4] or manipulation [5], [6]. Critically, each unique robot configuration requires addressing embodiment-and task-specific challenges in terms of control, sensing capabilities, perception, and planning. With changes in the number of propellers, structural materials, overall platform size, payloads, the onboard sensor suite, as well as the environment within which a system is expected to operate, autonomy design and optimization need to exploit high-end simulation toward a safer and faster path to resilient deployment.