Integrate FITRE (Purdue)

[podhrmic]

Faster and better training and re-training approach

Deliverables (Use branch subs/FITRE):

• FITRE trained NN for Inner loop LEC
• Jupyter NB to demo the integration and FITRE's performance improvements.

Tasks

• Integrate FITRE into PyTorch to enable seamless integration into the training workflow of inner-loop controller using stable-baselines.
• Show off FITRE's USPs by
  • note down the reduction in hyper-parameters and, if possible, use it to easily find better hyper-param values.
  • comparing training times with and without FITRE
  • using multiple initial conditions for training to improve the performance, while using the same timed budget as before.

Stretch

• Show off FITRE on a higher dimensional problem. For example, a multi-agent scenario with an appropriate objective: rejoin and complex maneuvers.
• FITRE trained NN for outer loop Autopilot