Multi-log grasping using reinforcement learning and virtual visual servoing