Throwing Manipulation: Design, Identification, and Learning-Based Control for a Novel Throwing End-Effector

Abstract

With the advent of Industry 4.0, the demands on industrial robots have expanded beyond simple pick-and-place tasks. Future smart factories require robots capable of a wide range of manipulation skills, including the ability to throw objects. This thesis investigates the design and optimization of a novel robotic end-effector that manipulates objects by enabling precise grasping and target throwing. Current robotic grippers primarily focus on grasping, while throwing is typically achieved through energy-intensive whole-arm movements. This approach is not only inefficient but also raises safety concerns. To address these limitations, this research proposes a versatile gripper that seamlessly integrates pick-and-place and pick-and-throw functionalities using stored elastic energy. The controlled release of this energy propels objects with accuracy, potentially exceeding the robot arm’s reachable workspace. Key contributions of this research include: • Novel End-Effector Design: The design of a new end-effector capable of performing pick, place, and throw actions without relying on whole-arm motion. This innovative design leverages stored elastic energy for throwing, thus enhancing efficiency and safety. • Physics-Based Model: A physics-based model that accurately correlates the stretch of an elastic band to the landing position of the thrown object. This model integrates the principles of rigid body dynamics to account for the object’s behaviour during projectile motion, making it essential for predicting and controlling the trajectory of thrown objects. • Parameter Identification: Implementation of a two-stage process to identify the parameters of the physics-based model. This process ensures that the model accurately reflects the behaviour of the end-effector. • Optimal Control Algorithms: Development of sophisticated control algorithms that enable the robot to throw objects to specific target locations with high precision. These algorithms calculate the optimal release point and force required for each throw • Data-Driven Residual Model: Development of a data-driven residual model to capture unmodeled dynamics and further improve throwing accuracy. This model uses machine learning techniques to refine predictions based on experimental data. • Experimental Validation: Conducting experiments to validate the effectiveness of the end-effector design and its control algorithms. These experiments demonstrate the practical viability and robustness of the proposed system. The thesis further explores the vast practical implications of throw manipulation. In warehouse logistics, this technology can significantly optimize sorting, packing, and distribution processes by enabling faster and more precise handling of items. In agriculture, it holds the potential to be used for harvesting, seeding, and the targeted application of resources, leading to increased efficiency and reduced labour costs. By combining optimization and learning-based approaches, this research provides a comprehensive framework for designing and optimizing robotic end-effectors for throwing manipulation. This interdisciplinary methodology enhances the versatility, adaptability, and performance of robots, ultimately improving efficiency, safety, and productivity in various industrial and operational settings. Keywords: Throwing manipulation, Gripper design, Trajectory optimization, Rigid body dynamics, Learning-based approaches.