a FPGA design of Dyna-Q Algorithm. LSI design Contest 2021.
We demonstrate our design in a simple maze search problem.
Dyna Q learning is among model-based planning reinforcement learning methods which learns Q function to estimate state-action value. Dyna Q learning is basically Q-Learning algorithm with extra planning steps.
Our design contains 3 main parts:
- Environment block.
- Training block.
- Q-table Block.
Experimental result (run on Altera DE2 board) shows that our FPGA design is correct.
See video demo at: Drive
Our source code is inspired by: