arnav dhamija shortstheory

Tasks

Integrating Fly-By-Logic to create trajectories in different environments and then applying L2F for fixing collisions
- Have a curriculum of multiple cases of obstacles and/or multiple drones.
- Give drones starting and ending velocities (might require a change in the environment since we're only using 3D position right now).
Changes to curriculum learning approach to become more flexible by training a 'Teacher' based on the features of the results of the Trainer. This could take very long to train.
- Alternatives to POMDP approach by observing slope of reward fxn here: https://arxiv.org/pdf/1707.00183.pdf. Appendix has simple algorithms, although could be tricky to implement because of rllib's distributed nature.
Have a variable time horizon of timesteps instead of a fixed number for both drones. Could be fixed by just using dummy states.
- Would current RL approach be applicable for variable time horizons?

	import os
	import glob
	import time

	base_dir = '/sys/bus/w1/devices/'
	device_folder = glob.glob(base_dir + '28*')[0]
	device_file = device_folder + '/w1_slave'

	def read_temp_raw():
	f = open(device_file, 'r')