Define/Finalize Our Action Space

[rlturkiye]

( continuous or discrete ? 2D movement or 3D movement ? if continuous which ranges? , if discrete which values to use) How action applied to drone ( it's acceleration , direct control of speed , turning angles etc)

[karascr]

We decided on four action spaces: 1- Roll, pitch, yaw, throttle 2- Control signal for each rotor 3- Velocity for each direction 4- Same as 3rd but without right and left and plus yaw

References:

• [https://www.mdpi.com/2504-446X/3/3/72](Deep Reinforcement Learning for Drone Delivery)
• [https://arxiv.org/pdf/1903.04628.pdf](Sim-to-(Multi)-Real: Transfer of Low-Level Robust Control Policies to Multiple Quadrotors)
• [https://arxiv.org/pdf/2103.02142.pdf](Learning to Fly—a Gym Environment with PyBullet Physics for Reinforcement Learning of Multi-agent Quadcopter Control)
• [https://arxiv.org/pdf/1804.04154.pdf](Reinforcement Learning for UAV Attitude Control)