Investigação de algoritmos de cinemática inversa para resolução de trajetórias em sistemas cobot pick-and-place baseados em ROS

Abstract

Industry 4.0 demands automation to enhance efficiency and precision, making collaborative robots essential due to their adaptability and integration with IoT and artificial intelligence. In this context, solving inverse kinematics in manipulators, particularly for pick-and-place tasks, is a significant challenge. This dissertation investigates numerical methods for solving inverse kinematics in a simulated environment using the GAZEBO software, organizing LEGO pieces with a Cobot integrated into ROS and computer vision via YOLO. Algorithms such as Gradient Descent (based on the pseudo-inverse of the Jacobian), which iteratively adjusts joint variables, and optimization methods like BFGS, L-BFGS-B, and Nelder-Mead, which use Hessian approximations to improve convergence, were compared. The results indicate that while analytical methods are superior in execution time, numerical methods offer greater flexibility and are easier to implement. However, their computational complexity and speed lag behind analytical approaches, making them a viable alternative for solving inverse kinematics problems in scenarios where analytical solutions are infeasible or nonexistent.