Reconfigurable Autonomous Surface Vehicles : perception and trajectory optimization algorithms

Abstract

Autonomous Surface Vehicles (ASV) are a highly active area of robotics with many ongoing projects in search and rescue, environmental surveying, monitoring, and beyond. There have been significant studies on ASVs in riverine, coastal, and sea environments, yet only limited research on urban waterways, one of the most busy and important water environments. This thesis presents an Urban Autonomy System that is able to meet the critical precision, real-time and other requirements that are unique to ASVs in urban waterways. LiDAR-based perception algorithms are presented to enable robust and precise obstacle avoidance and object pose estimation on the water. Additionally, operating ASVs in well-networked urban waterways creates many potential use cases for ASVs to serve as re-configurable urban infrastructure, but this necessitates developing novel multi-robot planners for urban ASV operations. Efficient sequential quadratic programming and real-time B-spline parameterized mixed-integer quadratic programming multi-ASV motion planners are presented respectively for formation changing and shapeshifting operations, enabling use cases such as ASV docking and bridge-building on water. These methods increase the potential of urban and non-urban ASVs in the field. The underlying planners in turn contribute to the motion planning and trajectory optimization toolbox for unmanned aerial vehicles (UAVs), self-driving cars, and other autonomous systems.