How mental imagery helps robot learning
Kristsana Seepanomwan · 2019
This work presents a computational model underlying learning in robots. It demonstrates the way in which during exploration, the robots not just gaining knowledge about movement primitives but also mental imagery capability. The target experiment involves permitting a humanoid robot to learn how to retrieve an out-of-reach object using a tool (action sequencing). Exhibiting exploratory behavior in the robot is possible through the help of a reinforcement learning technique and the dynamic movement primitive framework. The results show that the robot can complete a given action sequencing task in a reasonable period of times. In addition, by mean of planning, the robots able to examine actions' outcome using mental images alone. This could help reducing times and power consumption. Furthermore, two experiments which might be identical to the infant's age have been conducted and might be used to explain the characteristic of tool use development found in humans.