Combining backpropagation with Equilibrium Propagation to improve an Actor-Critic reinforcement learning framework