Reinforcement learning of 2-joint virtual arm reaching in computer model of sensorimotor cortex