Design strategies for controlling neuron-connected robots using reinforcement learning