Simultaneous control and human feedback in the training of a robotic agent with actor-critic reinforcement learning