Public defence in Automation Technology, M.Sc.(Tech.) Heikki Hyyti

The title of the thesis: Perception Systems for Autonomous Forest Machinery

Doctoral student: M.Sc.(Tech.) Heikki Hyyti
Opponent: Prof. Ola Ringdahl, Umeå Universitet, Sweden and Prof. Thierry Peynot, Queensland University of Technology (QUT), Australia
Custos: Prof. Arto Visala, Aalto University School of Electrical Engineering, Department of Electrical Engineering and Automation

In order for a robot to operate autonomously in a complex environment, it must perceive and understand its environment and its own actions there. If the environment is a forest and the robot is supposed to cut down trees, the robot must find the trees and free space to cut them down safely. In addition, the robot must know where and in which way it and its harvester head are located. Although this dissertation offers solutions to all these challenges, the robot cannot be assumed to work autonomously in all situations. That's why the operator should stay in the cabin to help the robot deal with problematic situations. In this case, the robot works semi-autonomously in cooperation with the operator, and the robot should also perceive the operator. 

This dissertation studies perception systems for autonomous forest machinery. These utilize machine vision, laser scanning, inertial, and positioning sensors. The work merges perception methods proposed in seven publications, in which the measurements from several sensors are fused together. The work involves developing a sense of balance for the robot, and a sense of the position of the boom and the tool. A three-dimensional perception has also been developed for the robot to detect trees and the ground surface underneath, to classify tree species, and also to identify spruce saplings among other vegetation. In addition, the pose of the operator's helmet has been measured so that an augmented reality user interface may be used in the cabin. 

All the solutions presented in the work seek to address a real existing problem. Several methods combine measurements from multiple sensors using probabilistic sensor fusion methods. These methods take into account both the uncertainty of the measurement and the measurement process, which is affected not only by the measuring device but also by the environment. All the presented methods and the required sensors have been implemented in prototype forest machines. The presented methods are suitable for real-time operation and they have been tested in a forest environment. 

The dissertation provides details about probabilistic sensor fusion and estimation methods such as Kalman and particle filtering, as well as their application to forest machinery. The work is intended for field robotics developers and researchers, as well as for those working with sensor systems for forest machines or other machinery.

Thesis available for public display 10 days prior to the defence at: https://aaltodoc.aalto.fi/doc_public/eonly/riiputus/

Contact information:

Doctoral theses in the School of Electrical Engineering: https://aaltodoc.aalto.fi/handle/123456789/53