Interaction Between Environment and Embodiment Using a Biped Robot

Recent research has claimed that incorporating bodily functions added to physical sensors, such as touching, smelling, and tasting will create AI that is able to classify and think based on its multisensory information. Using an apple as an example, AI recognizes an apple as a symbol, while humans recognize it based on multiple factors such as color, texture, and smell. Based on the idea that human body enables grasping the concept of things in detail, current studies have indicated that a physical body is essential for the establishment of advanced intelligence. This paper identifies differences between AI and human cognitive functions and discusses necessary means of recognizing AI. We reproduce the cognition of things by having the robot perform reinforcement learning of bipedal walking. In the experiment, the two robots reproduced object recognition by performing reinforcement learning of bipedal walking. By having virtual robots of the same mass and size as the actual robots learn to walk, we examine how their learning process, and the results differ between the robots with physical substance and simulators without it. Comparing physical and simulated robots allows us to examine the effects of robot’s physicality. We determine the environmental factors (friction, vibration, etc.) that vary the learning process and results, and suggest human-like cognitive functions the robots need to acquire.