Action estimation using a theory of mind as applied on the humanoid robot SURALP

Official URL: http://risc01.sabanciuniv.edu/record=b1534443 (Table of Contents)

Explanations regarding human consciousness have existed for a very long time. Theory of Mind (ToM) is one of the contemporary explanations for counciousness. This theory states that humans have functionalized brain parts for understanding beliefs and intentions of others. Humans have an inherent ability for making inferences on visual data once an acition is observed. Understanding/anticipating human actions based on visual data can be explained in context of ToM. It is proposed that a functionalized brain part is used for estimating intentions of others from observed movements of an actor. This functionalized part posses a Forward Model (FM) which simulates consequences of intentions. Simulated intentions are compared with observed movements to estimate the action of the actor. This thesis is based on implementation of such an action estimation model on a humanoid robot platform. A computational model for the part of the human brain which estimates intentions is needed to implement the model on a robotic platform. There is a proposed computational model in the literature for the part of the brain which estimates intentions. Model explains how a FM can be used along with a loop for action estimation by providing an algorithm. Motivation for such an implementation has two main reasons: To program a humanoid robot platform in such a way that it anticipates movements of the human actor to assist him/her, and a platform which can test ToM related to action estimation. In thesis the implementation is made on SURALP (Sabanci University ReseArch Labaratory Platform). Kinect is used for visual data input device. Various tests, which observe capabilities and limitations of the computational model, are completed with success.