Development and testing of advanced imaging methods for autonomous robot control

The availability of low-cost high-performance computing including General Purpose Graphic Processing units (GPGPU) and low-cost Lidar-based depth cameras offer the potential for the development of an inexpensive Lidar-based tracking system for use in autonomous robot control. Such algorithms can also include development of nonparametric 3D computer models of the robot environment. The 3D model would then enable the robot to physically interact with that environment in a way that would not be possible using alternative robot tracking techniques.

This work investigates the practical use and performance of a Lidar-based imaging and tracking algorithm operating at 30Hz and used to create a 3D model of the environment in real-time. The thesis performs a systematic investigation of the practical implementation and performance of such a real-time system using two low-cost test rigs.

These test rigs were developed and funded as part of the project by the author to investigate the practical aspects of implementing such systems on real hardware platforms using off-the-shelf Embedded computing and low-cost Depth-cameras, Radio-Control (RC) servos and sensors. From this work a lot of practical experience has been gained and some original contributions made in development of the tracking algorithms and specifically within the GPGPU codes. The test rigs also enabled the systematic development and performance evaluation of the tracking algorithms and the 3D model build process.

The performance of several documented Iterative Closest Point (ICP)-based tracking algorithms were evaluated for accuracy and performance and a new system developed here, using image features to improve tracking precision, was also tested.

The work also shows that Depth-camera based robot tracking is a very complex problem with no easy solutions and many issues relating to tracking performance were found during the development of this project. These are discussed in detail in the concluding sections of the report.