A benchmarking framework for motion planning algorithms in industrial manipulation

In the context of Industry 4.0, robotic automation continues to expand and plays a central role across a wide range of industrial applications. Although automation is already well established in production environments, the constant demand for new applications, together with the need to improve existing ones, requires a more accurate development of automated systems. In the field of robot manipulation, the efficiency of automated systems strongly depends on the capability to generate reliable robot trajectories. This task is addressed by motion planning, a field that offers a wide range of possibilities for further development. This thesis focuses on a comparative analysis of several relevant algorithms, highlighting their respective strengths and limitations. Testing and deploying motion planning algorithms in a consistent and comparable way is not straightforward. While the OMPL library and algorithms such as CHOMP and STOMP can rely on the almost ready-to-use structure provided by MoveIt2, others, such as TrajOpt and Descartes, required building a modular framework of classes and functions almost from scratch, which was implemented based on the Tesseract Robotics environment. A framework was therefore developed, able to simulate realistic industrial scenarios and compare different motion planning algorithms. All algorithms were evaluated on the same pick-and-place task, involving a UR5 manipulator, under different obstacle configurations and geometric constraints. The results show that both the outcome and the behavior of each algorithm strongly depend on the characteristics of the scene and on the tuning of its parameters. In particular, path length, trajectory smoothness, planning and execution time emerge as the features most sensitive to these variations, with these aspects often trading off against one another. These findings indicate that, across the wide range of possible industrial applications, no single algorithm or configuration performs optimally in every scenario, andatask-specifictuning of the motion planning system remains a necessary step for reliable real-world implementation.