Robotic manipulation of hand tools: estimation, planning, control and finger gaiting

Despite decades of research advancements, robotic hands are still incomparable with human hands in dexterity and versatility. Manipulation of hand tools, which usually requires sophisticated finger movements and precise controls, not only presents significant technical challenges but also carries a great potential for enabling the robot to assist humans in a wide range of tasks with tools.

Using screwdriving --- a complex and skillful manipulation task --- as a case study, this dissertation investigates the key components of dexterous robotic manipulation of hand tools, including estimation, planning, control, and finger gaiting. First, an extended Kalman filter (EKF) framework is developed to estimate the motion of the object in-hand, integrating vision, object dynamics and rolling kinematics. Next, a hybrid control scheme for robotic screwdriving is introduced, utilizing the dynamics of the screwdriver and the arm-hand system to achieve smooth screwing motion while maintaining a continual fastening wrench as the fingers roll on the handle. Finally, a dynamic finger gaiting technique is proposed, leveraging pivoting and adaptive contact forces to facilitate seamless transitions between grasp configurations. Additionally, a patch tree for hierarchical surface decomposition enhances grasp planning efficiency on curved surfaces, further advancing the adaptability and precision of robotic manipulation. Experiments with a Shadow Dexterous Hand demonstrated the accuracy and dexterity of the proposed framework in rolling-based manipulation tasks.

Committee: Dr. Yan-Bin Jia (major professor), Dr. Tichakorn Wongpiromsarn, Dr. Bowen Weng, Dr. Sourabh Bhattacharya, and Dr. Namrata Vaswani