Shared Control of Mobile Robots Using Model Predictive Control

Abstract

With the world constantly driving towards attaining complete autonomy, there is still a major question of safety when it comes to trusting a machine completely. Autonomous systems of today also do not have the ability to perform flawlessly in an environment that is cluttered and unstructured. This calls for the need of having a human operate the machine at all times either remotely via tele-operation methods or by being physically present alongside the machine. With tele-operation of remote systems, the cognitive load required from the human operator is high, while also the perception of the remote systems environment is low. This can cause many undesirable human errors causing damage to machinery. For example, tele-operating a forestry machine in a forest can be a very daunting task as there will be many trees and not all trees around the machine can be seen by the operator during remote tele-operation. With this in context, a few industries and sectors have now largely started research with using shared control methodologies to aid their machine in tele-operation tasks.

This thesis proposes a shared control methodology to provide a certain level of autonomy to the machine while still allowing the human operator to always be in control. The proposed methodology uses a Model predictive controller as the base controller to control the robot and perform obstacle avoidance tasks. The robot considered for implementation is a differential drive mobile robot, in specific the MiR 100 from Mobile Industrial Robots. The key motivation behind the thesis is to evaluate the performance of the shared control approach against a manual tele-operation task, to better understand the advantages and possible disadvantages of using a shared control strategy. The proposed strategy is implemented using the CasADi optimization toolbox on Matlab and tested through user testings. The results obtained from the user test prove that shared control can largely help in improving the safety of the system, but not so much with performance, at least not with the proposed methodology.