Foam-embedded Soft Roboitc Joint with Inverse Kinematic Modeling by Iterative Self-improving Learning
A novel soft robot arm design and a learning-based modeling paradigm inspired by active learning.
Authors: Anlun Huang*, Yongxi Cao*, Jiajie Guo*, Zhonggui Fang, Yinyin Su, Sicong Liu\(\S\), Juan Yi, Jian S, Dai, and Zheng Wang\(\S\)
* equal contribution
\(\S\) correspondence authors
(This work was submitted to IEEE Robotics and Automation Letters on 21st, Aug, 2023. Draft version: [pdf])
We propose a simple but effective structure in soft robot arm design to tackle the inherent high elasticity of existing solutions, parallelly embedding foams. The effectiveness is demonstrated with passive and active motion tests. Moreoever, we propose a pipeline to iteratively train inverse kinematic models of such soft arms. Inspired by the idea of active learning, our method resolves the difficulty of acquiring enough and high-quality data required for training an IK model. Experiments show that the modeling quality improves as we iterate the same training procedure threefold.