Exploring Robot Teleoperation in Virtual Reality

This thesis presents research on VR-based robot teleoperation with a focus on remote environment visualisation in virtual reality, the effects of remote environment reconstruction scale in virtual reality on the human-operator's ability to control the robot and human-operator's visual attention patterns when teleoperating a robot from virtual reality. A VR-based robot teleoperation framework was developed, it is compatible with various robotic systems and cameras, allowing for teleoperation and supervised control with any ROS-compatible robot and visualisation of the environment through any ROS-compatible RGB and RGBD cameras. The framework includes mapping, segmentation, tactile exploration, and non-physically demanding VR interface navigation and controls through any Unity-compatible VR headset and controllers or haptic devices. Point clouds are a common way to visualise remote environments in 3D, but they often have distortions and occlusions, making it difficult to accurately represent objects' textures. This can lead to poor decision-making during teleoperation if objects are inaccurately represented in the VR reconstruction. A study using an end-effector-mounted RGBD camera with OctoMap mapping of the remote environment was conducted to explore the remote environment with fewer point cloud distortions and occlusions while using a relatively small bandwidth. Additionally, a tactile exploration study proposed a novel method for visually presenting information about objects' materials in the VR interface, to improve the operator's decision-making and address the challenges of point cloud visualisation. Two studies have been conducted to understand the effect of virtual world dynamic scaling on teleoperation flow. The first study investigated the use of rate mode control with constant and variable mapping of the operator's joystick position to the speed (rate) of the robot's end-effector, depending on the virtual world scale. The results showed that variable mapping allowed participants to teleoperate the robot more effectively but at the cost of increased perceived workload. The second study compared how operators used a virtual world scale in supervised control, comparing the virtual world scale of participants at the beginning and end of a 3-day experiment. The results showed that as operators got better at the task they as a group used a different virtual world scale, and participants' prior video gaming experience also affected the virtual world scale chosen by operators. Similarly, the human-operator's visual attention study has investigated how their visual attention changes as they become better at teleoperating a robot using the framework. The results revealed the most important objects in the VR reconstructed remote environment as indicated by operators' visual attention patterns as well as their visual priorities shifts as they got better at teleoperating the robot. The study also demonstrated that operators’ prior video gaming experience affects their ability to teleoperate the robot and their visual attention behaviours.