Laser-based detection and tracking of moving obstacles to improve perception of unmammed ground vehicles

Laburpena

The goal of this thesis is to develop a system that improves the perception stage of heterogeneous Unmanned Ground Vehicles (UGVs) in order to achieve a robust navigation in terms of safety and energy saving in several real indoor and outdoor environments. The perception must deal with static and dynamic obstacles using heterogeneous sensors, such as, the odometry, LIght Detection And Ranging (LIDAR), Inertial Measurement Unit (IMU) and Global Positioning System (GPS), in order to collect the surrounding information with the highest precision, allowing to improve the planning and obstacle avoidance stages. To achieve this objective, a Dynamic Obstacles Mapping approach (DOMap) is proposed to obtain the local map with static and dynamic obstacles information. The proposal is based on the Bayesian Occupancy Filter (BOF), extended in order to not assuming discretized velocities. The velocities detection has been obtained using Optical Flow over a discretized grid of LIDAR measurements. In addition, the occlusions between obstacles had been handled and a multi-hypothesis tracking stage has been added, improving the robustness of the proposal. Therefore, the obtained map has information about occupancy and velocities of the surrounding obstacles (iDOMap). The proposal has been tested in simulated and real scenarios with different robotic platforms, including commercials ones and the developed PROPINA platfom to improve collaboration between humans and robots teams within ABSYNTHE project. Finally, LIDAR pose calibration and an improved odometry with IMU methods have been proposed.