dev roadmap

[hidal00p]

Define the core milestones to achieve in terms of program components and routines?

Define software stack

Composing a software stack is the most important part of this work, since my code will only be a user friendly wrapper around all the horsepower provided by NN engine, Gymnasium env, RL algorithms, hyperparam tuning.

• pybullet
• PID or betaflight interface?
• torch? rllib? optuna?

Concerns
Is it a feasible solution to train an RL agent using betaflight as a flight controller. Or is it an overhead?

Project workflow and user interaction mechanics

• How will user interface with it?
• How will the program interface with the available machine/cluster?
• What metric will be stored as a progress log?
• How will the training environment parameters be captured and stored? Do they contribute to reproducibility?

Which problem do I want to tackle?

• Do I actually want to work with lidar, or maybe some path following is sufficient?

[hidal00p]

This issue would be considered solved, once a suitable README exists, which addresses most of the bullet points.

[hidal00p]

This becomes a critical point again, since I need to think how do I formulate my RL problem?

[hidal00p]

I decided to stick with the obstacle avoidance problem. This is probably the simplest approach, which will make use of the CtrlAviary environment.

Based on the this I am going to formulate the next work pkgs.