[RL] Action Space
ugurkanates opened this issue · comments
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)
3- Velocity for each direction - > 2 Versions . Join = thread join , waits (duration) , // no join = ORIENTATION = 3 used,(sim speed 5)
z Function to use
yawRate = calculated