Jayven Cachola and Penny Silliman
Implemented tf-socket agent which uses TensorFlow DL networks to decide actions to take in Blind FSM. It does not do Deep Q Learning at this point (July 2024). See the paper submitted to ICAART 2025.
Jayven began work to support non-goal sensors (e.g., odd/even) but we got stuck
because it's unclear how best to do predictions with the DL network. Also, the
scope of this project exceeded our remaining time. So, it's currently incomplete.
Penny used this branch to experiment with different DL network designs.
No need to merge.
We added e-Greedy to the agent. We will likely merge this later this summer
Jayven experimented with an alternative approach to truncating predicted action
sequences. See Journal for details. This had no significant effect so it was not merged.