Deformation-based multimodal perception for soft pneumatic robots

Abstract

Soft robots made of highly compliant and elastic materials have attracted attention in the robotics field. Compared to the rigid-body robotic systems composed of rigid links, soft robots exhibit more advantages in terms of complicated tasks requiring safety and interacting with humans and unstructured environments. Because soft robots have adaptability and flexibility resulting from inherited compliance. However, soft robots still require sensory feedback. It can enable soft robots to accomplish complicated tasks, like accurate closed-loop control, by understanding their surroundings and monitoring their real-time self-states.

Although various previous research works have recently been reported to endow soft robots with single-type robotic perception by accompanying the internal and external sensors, the high-level multimodal perception for the soft robots by embedded sensors is still challenging. (1) Existing self-sensing soft robots integrate sensors into soft actuators where the heterogeneous stiffness between the sensors and actuators leads to complicated fabrication and even attenuates the soft robot’s compliance. (2)The modeling of soft robots’ state estimation to achieve multimodal sensing is still difficult due to the soft material’s inherited nonlinearity and strong decoupling between sensing modalities. (3) Construction of the robust perception strategy for soft robots is required to enable them to complete tasks still when partial subsystem failure occurs. (4) Once the sensing abilities are developed for soft robots, the high-level closed-loop control still needs to be investigated for creating smart soft agents.

This thesis aims to enable soft pneumatic robots with high-level multimodal perception with embedded internal sensing methods. To this end, the deformation-based sensing scheme for soft pneumatic robots is proposed using discrepancy characteristics between the heterogeneous sensing states of the soft actuator when stimuli are exerted. Based on this sensing method, both internal state perception (proprioception) and external interaction estimation (exteroception) are implemented on soft pneumatic robotic systems with different types and various degrees of freedom (DOFs). Then, the multimodal sensing scheme is implemented in the 3-DOF soft gripper and 3-DOF soft parallel joint. They are driven by self-sensing soft pneumatic actuators with origami structures to achieve the one-dimensional and three-dimensional position and force perception, respectively. These self-sensing soft actuators integrate two sensors for the heterogeneous modalities: a pressure sensor to measure the internal pressure of the actuator and a displacement sensor to measure the primary deformation of the actuator. Finally, a high-dimensional soft modularized robotic platform consisting of a 1-DOF soft gripper and a three-segment soft manipulator (connecting the three soft parallel joints serially), driven by self-sensing soft origami actuators. And a robust networked proprioception scheme allowing for sensor failure explored for the soft manipulator to to investigate the closed-loop control, validated by the three-dimensional position control applications.

The presented conceptual, methodological, and experimental efforts can be extended to other large families of intelligent soft pneumatic robots. They may also be fully leveraged in various real-world tasks involving exploring unstructured environments and human interactive collaborations.