Soft Hand Kinematics
Learning Kinematics of a Soft Hand
Intro
The RBO hand 3 [1] is a soft robotic hand developed at Robotics and Biology Laboratory (RBO) at Technical University of Berlin. It is a highly compliant hand designed for dexterous grasping. The inherent compliance of the fingers makes it hard to model them analytically unlike their rigid links counterparts. Therefore, in this project, we investigate efficient data-driven approach to learn the forward and inverse kinematic models of the hand to control and visualize the hand in free space.
We used a small feed-forward network to model the pose of fingertips as a function of airmass. In addition, palm and T{1-3} bellow actuators are modeled as revolut joints. The ring and little finger, and thumb form a chain of connected joints with their respective bellows. The final finger pose is computed by taking into account airmass in each finger compartments along with palm and thumb bellows (T1,T2,T3).
Kinematics-Driven Fingertip Prediction App
Building upon models I developed in my project work, I created a Flutter-based [3] app during my free time to predict fingertip positions from airmass data and vice versa. By integrating forward and inverse kinematics models, this app provides precise and real-time predictions. The project allowed me to repurpose and refine my earlier research while expanding my knowledge in mobile app development.
Team
Sumit Patidar, Adrian Sieler (Mentor)
References
[3] Flutter - build apps for any screen. Available at: https://flutter.dev/ (Accessed: 12 December 2024).