Details about observation space #96
Description
Hi,
I am looking to collect data from PyFlyt and use the observations for a downstream task. I am using the fixed-wing waypoints task with a context length of 2. I just wanted confirmation on whether my understanding of the observation state is correct or not. Each observation is has 29 columns and below are the mappings I have gotten from chatgpt but would like someone to confirm.
Thanks
--- OBSERVATION SPACE MAPPING (29 Columns) ---
- Physics State (0-20)
Indices 0-2: Angular Velocity (Body Frame)
0: AngVel_X (Roll Rate)
1: AngVel_Y (Pitch Rate)
2: AngVel_Z (Yaw Rate)
Indices 3-6: Orientation (Quaternion - PyBullet Standard)
3: Quat_X
4: Quat_Y
5: Quat_Z
6: Quat_W <-- SCALAR COMPONENT IS LAST
Indices 7-9: Linear Velocity (Body Frame)
7: LinVel_X
8: LinVel_Y
9: LinVel_Z
Indices 10-12: Global Position (Inertial Frame)
10: Pos_X
11: Pos_Y
12: Pos_Z
Indices 13-16: Last Action (Control Inputs)
13: Act_Roll
14: Act_Pitch
15: Act_Yaw
16: Act_Thrust
Indices 17-20: Auxiliary State
17: Aux_Thrust (Previous Thrust)
18: Aux_0 (Unused/Pad)
19: Aux_0 (Unused/Pad)
20: Aux_0 (Unused/Pad)
- Context / Targets (21-28)
Indices 21-24: Next Target (Immediate Goal)
21: Rel_Dist_X
22: Rel_Dist_Y
23: Rel_Dist_Z
24: Rel_Yaw <-- Non-zero in your data (Heading to target)
Indices 25-28: Future Target (Lookahead)
25: Rel_Dist_X
26: Rel_Dist_Y
27: Rel_Dist_Z
28: Rel_Yaw