Low-Bandwidth Three Dimensional Mapping and Latency Reducing Model Prediction to Improve Teleoperation of Robotic Vehicles
Type of Degreethesis
MetadataShow full item record
This thesis uses novel three dimensional sensors like the Microsoft Kinect  and the Asus Xtion Pro Live  to generate three dimensional environments and use the recon- structed environment with a predictive model in order to assist the teleoperation of mobile vehicles. Ultimately this work would be applicable to any teleoperated vehicle equipped with sensors providing three dimensional data of the environment, such as an automated ATV with a stereo vision system or a Velodyne LiDAR  system. The challenges related to utilizing dense three dimensional data in a way that is practical for teleoperation scenar- ios are identified, and solutions are proposed and implemented. To simplify the approach, the problem is split into three smaller tasks: three dimensional mapping, teleoperation and telemetry visualization, and latency reduction techniques. The three dimensional mapping pertains to using the three dimensional sensor data in concert with the mobile vehicle nav- igation solution to generate a three dimensional map of the environment in real-time. The resulting map must be efficiently sent to the teleoperator and visualized in the teleoperation and telemetry visualization section of the thesis. Additionally, latency greatly reduces the teleoperator’s ability to drive the vehicle, so methods for reducing the perceived latency are investigated, including using a vehicle model to simulate the vehicle motion in the absence of timely telemetry updates. It is shown that existing mapping techniques can be used effi- ciently and effective to aid teleoperation, even in low bandwidth environments. Experimental results show that by giving the teleoperator three dimensional information about the envi- ronment, the teleoperator can more successfully navigate tight obstacles and reduce impacts with the environment. Finally, experiments are conducted that show having a prediction of the vehicle motion based on user input can improve teleoperation in high latency situations.