Hi all,
I'm currently doing some testing of using RealFlight as a simulation environment for research into reinforcement learning for aerobatic flight control of UAVs. However, there are a number of limitations that currently limit its use. The primary issue is one of speed. Currently, from what I can see, the physics engine can only be changed to run at 150% real-time. This is currently too slow for reinforcement learning use, with the need to run thousands of episodes.
Does anyone know of any way of running the physics faster than this? I think RealFlight/FlightAxis could be a really promising tool for my research if I could find a way of overcoming this.
I'm currently doing some testing of using RealFlight as a simulation environment for research into reinforcement learning for aerobatic flight control of UAVs. However, there are a number of limitations that currently limit its use. The primary issue is one of speed. Currently, from what I can see, the physics engine can only be changed to run at 150% real-time. This is currently too slow for reinforcement learning use, with the need to run thousands of episodes.
Does anyone know of any way of running the physics faster than this? I think RealFlight/FlightAxis could be a really promising tool for my research if I could find a way of overcoming this.