Soft pneumatic robots with bodily awareness

Abstract

Over the past decade, roboticists have been experimenting a huge variety of approaches to enable future robotic systems to implement increasingly complex tasks and interact with humans in daily life. Soft robots made out of soft materials or compliant mechanisms are promising candidates to operate in physically-proximal human-robot coexisting environment. As distinguishing features compared to the classic rigid robots built with metals, soft robots have lower hardness and mass, which ensures intrinsically safe interactions between humans and robot. Also, attributing to the mechanical compliance, soft robots can easily achieve hardware intelligence without relying much on low-level sensory feedbacks and obviate complex control problems. The passive compliance of soft robot is a powerful advantage can be leveraged for tackling many typical robotic problems, but not a panacea. To push the boundaries of soft robots for dealing with more challenging tasks, sensing elements and software intelligence are required to endow soft systems with capabilities of perceiving their own body state and feeling the external world. To date, a variety of advanced technologies for constructing perceptive soft robots have been studied, while there is still a gap to utilize them in practical applications due to the challenges in sensing system development, modelling and data interpretation: (1) most existing soft sensing systems are based on dedicated smart material, which usually require complex fabrication process and behave with nonlinearity, hysteresis, and viscoelastic effect; (2) reconstruction of multi-DoF deformation is still an intractable problem due to the complex soft-body dynamics and the mechanical coupling between the integrated sensors and the robot body; (3) data-driven models that can extract useful information from raw sensor data for high-level perception are under explored. This thesis aims to endow soft pneumatic robots with the abilities to sense their own states, the interactions, and the environment. A novel structure-based bodily awareness concept was proposed for soft pneumatic robots, by using elastic chamber as sensing body and encoding the internal and external mechanical information to unified pressure signals. Based on the proposed concept, both proprioceptive and exteroceptive modalities were explored and experimented on soft pneumatic systems with different morphologies. Firstly, a modeling-based approach was proposed to enable a single-DoF soft-rigid hybrid gripper actuated by antagonistic bellows chambers with multimodal sensing of actuator movements and grasping force. In addition, proprioceptive sensing scheme was proposed for multi-DoF soft continuum joint by configuring sensing network with distributed bellows chambers and using learning-based approach to generate the kinematic model from the unified pressure signals. The performance of the multi-DoF soft system under sensor network failures was discussed and both hardware and software level solutions were explored for graceful degradation. Lastly, high-level perception of the environmental properties was achieved with preliminary studies of soft pneumatic system with proprioceptive and tactile sense. The soft-rigid hybrid gripper prototype was used to demonstrated two interesting tasks: (1) reconstruction of the contact surface geometry by mechanical scanning motion; (2) identifying soft objects with different deformation properties by pinch grasping.