Implementation of feature based object identification in Bipedal walking robot

Abstract

A bipedal walking robot resembles human. They are specifically programmed to perform some specific tasks. The field of Robotics is growing rapidly to serve military and commercial applications. Bipedal are capable of doing almost all crucial and critical tasks which are dangerous for humans. To fulfill this aim the bipedal should have a vision system. This system would help to identify the objects and the bipedal controller would be able to take actions. The present work deals with a vision based navigation (VBN) of bipedal. The bipedal identifies the object by using SURF algorithm. The action strategy of navigation is with the bipedal controller which uses Q-learning RL algorithm in dynamic environment. Bipedal identifies the object depending on the objective of its design or objects stored in database. This feature based Q learning RL algorithm helps in reducing the number of states values and also help in sharing and transferring the knowledge from one RL agent to another RL agent. Also useful for obstacle avoidance and identifying the dangerous objects while navigating.