edgorman / pencil-landing-simulator Goto Github PK

Make the file structure similar in style to URL Analyser

This also includes adding argparse for user input via command line,
and simplifying the environment.yaml file

Maybe one at sea where the angle moves with some sin wave

Using the config file to specify them

Need to decide how this links with the model saving stuff, probably just another folder within 'ppo' like 'earth'

https://github.com/edgorman/Pencil-Landing-Simulator/blob/main/PLSimulator/environments/environment.py#L108

Model waves using simple sine wave
Platform angle is the tangent to the wave at a fixed position along wave

Make sure step function reports the angle as relative to angle of agent

Besides PPO lol

The base environment should control all physics processes

Child environments can add their own processes that affect the forces applied to the agent (e.g. air resistance)

Need to pass in list of entities acting in environment, alongside the forces acting on them
-> This requires a new class of 'entities' that will be rendered, agents/landing platforms/obstacles are subclasses of this
-> Each one should have code dictating how it moves (e.g. platforms are fixed, agents will move)

Move spaceenv step code to baseenv, rename planetenv to gravityenv and create a new env called dragenv which inherits from gravityenv

Also fix drag calculation

Currently the agent likes to stay at the top of the screen and fire the engine until it has no fuel

This is likely because we reward a low velocity, but we should only give this reward when close to the goal

Put similar instructions in the readme

and copy the action file from url analyser

Probably add these to the output model folder

Randomly alter the agent starting conditions when environment.reset() is called, including position/velocity/angle.

Make the reward system dependent on the environment config given (e.g. reward for firing engine based on unit distance to pad)