In this paper Kernel-Based Reinforcement Learning (2002), a classical kernel-based method was demonstrated for Reinforcement Learning , which indicates that classical research in this direction is well underway. Furthermore, in supervised learning, it has been observed that Quantum Kernels offer advantages over QVCs (Guaranteed optimal measurement with kernels, no barren plateaus, kernels not dependent on ansätze, etc.). Could we also consider applying Quantum Kernel-Based Reinforcement Learning (instead of using a Variational Quantum Policy Circuit, employing a kernel method)?
Edit: Here is the paper that demonstrates the advantages over VQC in supervised learning: arXiv:2101.11020
If yes, can you guys link some papers?
