Refactor drone environments and merge back into one env

[stmio]

Major changes to drone environment to make development for sim2real easier:

• Merge swarm and racing envs back into one
• Split task logic into a separate file
• Separate rendering logic from physics
• Improve overall code quality

[Kinvert]

I made an env based off your guys stuff FYI:

#372

[stmio]

@Kinvert - looks awesome! Maybe we should work (at some point in the future) on adding this to the main "drone" environment as a task. I think it would be super cool to have it all in one place, where you can pick and choose which tasks to train on given a scenario. :)

[Kinvert]

Yep this is something I wanted to do after the hackathon but ran out of time. I had some small changes to that physics h file. Your envs might benefit actually from some sort of PID. iirc they seemed to be just training off the P portion.

The Drone Delivery task could be simplified greatly by getting rid of the hidden_pos etc. This was done since we wanted the complexity in PufferLib and not have to manually move around the actual target position and velocity in Isaac Sim.

[FinlaySanders]

I think the "I" is handled by the LSTM?

[Kinvert]

Not sure they seemed to wobble to me, but I was in a hackathon and didn't have time to be thorough lol.

[stmio]

Some things we are working on for the next version:

• Proper sim2real results! (dev for this is here https://github.com/tensaur/drone)
• Better logging and perf metrics
• Policy struggles with learning both static/dynamic tasks simultaneously. This is improved in this version, but there is still room for improvement.

[FinlaySanders]

Worth noting that the current perf metric needs some work - we will make this nice soon.

[FinlaySanders]

Rough sim2real todo:

• Make physical constants in dronelib.h match real crazyflie drone
• Add observation noise

[Kinvert]

Consider some actions noise too, latency, etc. Or maybe it was already there and I forgot. Physics noise.

Also one thing I don't remember seeing, could have forgot or missed it, was how they can induce yaw by changing the 4 throttles. Was this accounting for torque of the blades? Gyroscopic effects of the blades?

[FinlaySanders]

I think these bases are covered already but will double check :). I suspect for some of the aerodynamic forces there will be a tradeoff. ie. can we measure the constants accurately enough that including them will be worth it (or find them elsewhere)