Sensor to motion coordinate frame calibration in robotic vehicles

• Autor(a) principal: Gomes, Manuel Alberto Silva
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/34368
• topic: Extrinsic calibration; Odometry; Atomic transformations; Mobile robots; ROS

Extrinsic sensor calibration is crucial in operating a robot and is needed for several functionalities, such as sensor data fusion, 3D reconstruction, and SLAM. Although various state-of-the-art approaches exist to tackle this problem, the majority focus on a pairwise calibration, i.e., the calibration between two sensors. In a system with more than a pair of sensors, the number of calibrations increases, creating a tedious and unoptimised procedure. To solve this, the Laboratório de Automação e Robótica at the University of Aveiro developed a calibration tool named ATOM. This approach formulates the calibration problem as an extended optimisation problem in which the poses of the sensors and the poses of the calibration pattern are estimated. These poses are retrieved by the combination of geometric transformations, named atomic transformations, due to their indivisibility. This allows for a generalisation of the calibration problem, making the approach viable for every robotic system and every multiple configurations. Calibration w.r.t. the motion coordinate frame is a sensor to frame calibration w.r.t. the robot frame. This robot frame is moving w.r.t. the odom frame, which is a world-fixed frame computed based on an odometry source. This calibration allows for a near-global placement of the robot’s sensors, which allows for object and environment positioning w.r.t. the robot. The author argues that ATOM’s mathematical formulation can be used in this type of calibration due to its broad and general nature. Therefore, this dissertation focus on developing tools to test the proposed approach in simulated and real environments. Results show that a successful proof of concept of this approach was proven in a simulated environment, achieving results comparable with normal calibration in ATOM. On the real environment, result were unsatisfactory due to external reason, such as uncalibrated odometry parameters.