Self-Body Image Acquisition and Posture Generation with Redundancy using Musculoskeletal Humanoid Shoulder Complex for Object Manipulation (2411.06320v1)

Abstract: We proposed a method for learning the actual body image of a musculoskeletal humanoid for posture generation and object manipulation using inverse kinematics with redundancy in the shoulder complex. The effectiveness of this method was confirmed by realizing automobile steering wheel operation. The shoulder complex has a scapula that glides over the rib cage and an open spherical joint, and is supported by numerous muscle groups, enabling a wide range of motion. As a development of the human mimetic shoulder complex, we have increased the muscle redundancy by implementing deep muscles and stabilize the joint drive. As a posture generation method to utilize the joint redundancy of the shoulder complex, we consider inverse kinematics based on the scapular drive strategy suggested by the scapulohumeral rhythm of the human body. In order to control a complex robot imitating a human body, it is essential to learn its own body image, but it is difficult to know its own state accurately due to its deformation which is difficult to measure. To solve this problem, we developed a method to acquire a self-body image that can be updated appropriately by recognizing the hand position relative to an object for the purpose of object manipulation. We apply the above methods to a full-body musculoskeletal humanoid, Kengoro, and confirm its effectiveness by conducting an experiment to operate a car steering wheel, which requires the appropriate use of both arms.

• The paper demonstrates that integrating enhanced muscle actuators in the humanoid shoulder complex improves joint stabilization and posture control.
• The authors leverage a calculated scapulohumeral rhythm with a 1:2.7 ratio to generate stable postures through inverse kinematics.
• Experimental results validate the approach by showing effective shoulder actuation in tasks like vehicle steering for precise object manipulation.

An Expert Analysis of Self-Body Image Acquisition and Posture Generation for Musculoskeletal Humanoids

This paper presents a comprehensive approach for improving posture generation and object manipulation in musculoskeletal humanoid robots, with a particular focus on the humanoid shoulder complex. At its core, the paper introduces methods for learning self-body image through inverse kinematics and enhanced hardware design inspired by human biomechanics.

Musculoskeletal Humanoid Design and Implementation

The authors have developed a musculoskeletal humanoid shoulder complex that closely mimics human anatomy. This design includes an integrated scapular drive mechanism to maximize joint redundancy, which is a pivotal attribute for achieving a broader range of motion. The hardware enhancement is centered on stabilizing the shoulder complex through increased muscle redundancy. This is accomplished by implementing deep muscle structures that aid in joint stabilization—a facet historically overlooked due to spatial constraints.

Two significant advancements are articulated:

1. Enhanced Scapular Stabilization: The introduction of additional muscle actuators mimicking the trapezius muscle enhances scapular mobility and stability. This setup expands upon the conventional four-muscle actuation system by integrating two extra lower trapezius actuators.
2. Joint Stability via Rotator Cuff Implementation: The design embeds a rotator cuff-like structure, augmenting existing shallow muscle actuators with deeper layers. This arrangement ensures resilience against dislocation and maintains stability for the spherical glenohumeral joint.

The authors have implemented these advancements on the Kengoro humanoid, conducting basic validation experiments to demonstrate effective scapular and glenohumeral joint actuation.

Posture Generation and Inverse Kinematics

The paper proposes a novel method for generating postures in musculoskeletal humanoids that leverages the scapulohumeral rhythm inherent in human physiology. Typically, musculoskeletal robots are plagued by the complexities of joint and muscle redundancy, leading to indeterminate joint angles for a given task. In contrast, the proposed method ensures stable posture by maintaining a calculated relationship between the scapulothoracic and glenohumeral angles, thereby reducing the risk of joint dislocation.

This approach to posture generation divides the shoulder complex operation into distinct phases:

• Initial posture-based inverse kinematics calculations are performed.
• Scapula angles are adjusted using a constant derived from the scapulohumeral rhythm, specifically a $1:2.7$ ratio, for realigning joint mechanics.

Self-Body Image Learning for Object Manipulation

To address obstacles associated with real-time posture correction, the authors present a method for self-body image acquisition, crucial for precise object manipulation. Kengoro's posture is initialized to a user-defined stance that minimally burdens the robot's musculature. The method employs a vision-based system for estimating hand-object relative positions, serving as a reference for calibrating joint angles via inverse kinematics applicable to unobserved postural changes.

Experimentation and Results

A key aspect of the research validation is the robotics experiment involving vehicle steering wheel operation. Without prior learning, the humanoid struggles due to the inability to account for internal deformations and excessive muscle redundancies. Post self-body image acquisition, Kengoro accomplishes the steering task effectively, demonstrating the robustness of the proposed methodologies.

Implications and Future Directions

The implications of this research are twofold: practically, it facilitates deploying humanoid robots in complex manipulation tasks without significant environmental modifications; theoretically, it offers insights into human biomechanics replication in robotic systems.

Future explorations may explore:

• Integration of non-muscle structures to synergize with muscle actuators for enhanced joint stabilization and energy efficiency.
• Progressive self-body recognition algorithms utilizing multi-sensory inputs.
• Application of the methods to dynamic tasks where object-relative positioning fluctuates, requiring adaptive self-referencing mechanisms.

This paper represents a significant contribution to the field of humanoid robotics, addressing critical challenges in biomechanical mimicry and control strategies through inventive design and adaptive posture management techniques.