Exploiting the Intrinsic Embodied Dynamics for Adaptive Robotic Behaviors

Abstract

Robots have been evolving from tackling the traditional basic pick-and-place tasks to more complicated tasks such as general-purpose manipulation. Inspired by biological systems, current roboticists have utilized the concept of embodied intelligence for the development of adaptive robotic systems. This concept hypothesizes that intelligent behaviors result from the strict coupling of the physical interactions of the brain, the body and the environment. By utilizing the system-environment interaction, the "brainless" robotic systems have emerged with a reduced computation load on the neural process. This technique enables compliant interactions with the environment, allowing for more adaptive behaviors in complex and unstructured environments. This thesis explores how the intrinsic embodied dynamics can be utilized for the design of adaptive robotic systems with behavioral diversity. On the one hand, to occupy a specific ecological niche the robot is required to be designed with specific characteristics that maintain its situatedness in the environment. Under this circumstance, environmental changes can cause the system to break down. On the other hand, creating flexible robots that adapt to environmental changes will undoubtedly impose a computational load on neural processing. This thesis contributes to addressing this "Specific-Flexible" dilemma by introducing the intrinsic embodied dynamics in the robot design. Integrating materials and morphology in the mechanical design enables adaptive behaviors while leveraging the redundancy principle to allow for behavioral diversity. To facilitate self-organization and behavioral emergence in system-environment interactions, proper controllers should be designed to coordinate these embodied dynamics. This thesis is divided into three main parts and three main hypotheses have been raised and several scientific researches have been carried out for their validation. In Part I, I investigate the variable stiffness systems. I integrate material autonomy and several morphological designs into end-effectors to form a redundancy design. Both a particle-jamming soft finger and a universal gripper are developed to understand how the proper embodied dynamics design can enable robots to achieve complex and universal manipulation tasks. The incorporation of self-healing into the universal gripper design expands the robotic behavioral diversity and contributes to the exploitation of material intelligence in pick-and-place benchmark tasks. In addition, I propose a reduced-order modeling framework to capture the essence of hybrid soft-rigid systems in keystroke applications. In Part II, I investigate the self-organized control in human motor movements. I explore the emergence of self-organization and adaptive pattern transition behaviors within a piano-playing context. The human upper-limb coordinate is studied through real-world physiological experiments and simulations. For a simulated pianist, the coupling between the controller, body, and environment within the embodied intelligence framework is understood through the use of the central pattern generator (CPG), the Hill-type muscle model, and the piano. The joint trajectory pattern is characterized by repetitive arm swing behaviors and keystroke actions. This system is analyzed to understand how humans utilize passivity and morphological intelligence to reduce the neural processing load. In Part III. I study the adaptive robot in a human-centered application. I propose a human-robot cooperation framework where the robot adaptively teams with its human partner to fulfill a piano-playing task. The self-organized control of the robotic behaviors assists in understanding how decentralized control enables more adaptive behaviors in human-centered applications. Bidirectional communication relies on non-verbal cues, where the human and robot are not only a leader-follower dynamic but a reciprocal communication relationship. The robot maintains real-time adaptation to the human player and facilitates accurate timing and temporal alignment. The robot's fast-reacting behavior aligns with the concept of embodied intelligence. Lastly, this thesis concludes with a discussion of its contributions and outlines the perspectives and scopes of future research directions. In summary, the works in the thesis focus on utilizing the intrinsic embodied dynamics for the designs of adaptive robots. By utilizing proper material autonomy and morphological designs of the physical body, the robots achieve adaptivity and behavioral diversity in the system-environment interactions. The neural processing load of the robots can be significantly reduced while maintaining behavioral diversity. Thanks to the intrinsic embodied dynamics within the morphological design and self-organized control, robotic systems demonstrate abundant adaptive behaviors in future complex and unstructured applications.