Making a robot explain its decisions

Martin Cvetkov (2017) Making a robot explain its decisions. MSc thesis.

Preview

PDF Download (1606Kb)

Abstract

This master thesis is concerned with making robots more understandable to the user and broader audience. More exactly, the problem that this thesis tackles is to develop a way for robots to explain their actions in order to make it easier for engineers, users and for the audience to understand what the robot is actually trying to achieve. Another problem that this thesis tackles is that the tasks of the robots may not be precisely defined upfront. In such cases, tasks cannot be solved by a fixed, predefined task plan. Thus, I also show how to enable the user to create custom scenarios that the robot successfully resolves. These two problems are resolved by using planning. More specifically, the initial state of the robot and the goals that need to be achieved are defined by the user. After that, the robot, using the goal regression algorithm, creates a plan in order to achieve the goals. Each action of the plan is explained using an explanation algorithm. The algorithm takes into account the immediate and indirect goals the actions achieve. As a result of this master thesis, I developed a program that simulates and can control a robot with six degrees of freedom. The program creates dynamic plans for user defined goals. Also, the program enables the robot to explain each action it makes to the user, using the speakers.

Actions (login required)