Work

学習と冗長性を活用したハードウェア損傷と変形特性を補償するソフトロボットの実現

Year: 2025
Type: dissertation
Abstract: Soft robots have shown significant promise in performing interactive tasks challenging for traditional rigid robots.Their compliance offers substantial advantages in adapting to their surrounding environments, which enables a wide range of applications.These applications include operations in unstructured or human-inaccessible environments.On the other hand, the deformability of soft robots (i.e., response to given control inputs) is highly unstable.In particular, their vulnerability to damage poses a significant challenge, as even minor damage can drastically alter their behavior and impair functionality.Although many studies have addressed these issues, existing methods have yet to achieve a comprehensive ability to compensate for changes in deformability, including hardware damage, without requiring downtime or external intervention.For practical applications, soft robots must autonomously and continuously maintain functionality despite variations in deformability.Without this capability, soft robots may fail to operate reliably in unstructured environments, unstable conditions, or areas inaccessible to humans.Addressing this challenge can improve their reliability and enable applications that are challenging for rigid robots.Inspired by the redundancy in humans, which allows for the remarkable preservation of sensorimotor function despite fatigue or injury, this thesis addresses the aforementioned challenge by utilizing learning (neural networks) and redundancy.The objective is to enhance the ability of soft robots to autonomously compensate for hardware damage and changes in deformability.To achieve this objective, this thesis presents four interrelated studies.Chapter 2 addresses deformability compensation through control and proposes a learningbased controller that enables autonomous and rapid adaptation to changes in the deformability of a single soft actuator.From Chapter 3, this thesis addresses a key limitation of controller-only compensation, its failure under severe actuator damage, by introducing functional redundancy.By integrating actuator redundancy with a learning-based controller, compensation for the severe deformability variation, including those caused by hardware damage, is achieved.Chapter 4 focuses on soft sensors, which are critical for deformability compensation while maintaining compliance.To address the vulnerability of soft sensors, a learning-based proprioception framework is proposed for redundant soft sensors, which compensates for sensor damage (degradation).Finally, Chapter 5 integrates the redundant actuator controller and the proprioception framework, enabling a soft robot with embedded soft sensors to autonomously adapt to deformability changes, specifically those resulting from damage.Through these four steps, to the best of the author's knowledge, this
Language: English
Open Access status: green
OpenAlex ID: W7144798533
Cites: 267
Cited by:
Related to: 10