Antagonist Inhibition Control in Redundant Tendon-driven Structures Based on Human Reciprocal Innervation for Wide Range Limb Motion of Musculoskeletal Humanoids (2409.00705v1)

Abstract: The body structure of an anatomically correct tendon-driven musculoskeletal humanoid is complex, and the difference between its geometric model and the actual robot is very large because expressing the complex routes of tendon wires in a geometric model is very difficult. If we move a tendon-driven musculoskeletal humanoid by the tendon wire lengths of the geometric model, unintended muscle tension and slack will emerge. In some cases, this can lead to the wreckage of the actual robot. To solve this problem, we focused on reciprocal innervation in the human nervous system, and then implemented antagonist inhibition control (AIC) based on the reflex. This control makes it possible to avoid unnecessary internal muscle tension and slack of tendon wires caused by model error, and to perform wide range motion safely for a long time. To verify its effectiveness, we applied AIC to the upper limb of the tendon-driven musculoskeletal humanoid, Kengoro, and succeeded in dangling for 14 minutes and doing pull-ups.

• The paper introduces AIC, which dynamically adjusts muscle stiffness using reciprocal innervation to minimize antagonistic tension in humanoid systems.
• Experimental tests on the Kengoro platform demonstrate AIC's advantages over traditional Muscle Stiffness Control by handling geometric model errors.
• AIC advances biologically inspired robotics by enabling robust, safe, and efficient wide-range motions in tendon-driven musculoskeletal humanoids.

Antagonist Inhibition Control for Wide Range Motion in Musculoskeletal Humanoids

This paper presents innovative research focusing on the development and validation of a control mechanism termed Antagonist Inhibition Control (AIC) tailored for tendon-driven musculoskeletal humanoids. This control paradigm is inspired by the human nervous system's reciprocal innervation process, which efficiently coordinates agonist and antagonist muscle activities to facilitate smooth motion while minimizing internal muscle tension and potential damage.

The principal challenge tackled in this paper arises from the complexities inherent in creating a highly accurate geometric model for musculoskeletal humanoids. The geometric discrepancies between a model and its physical counterpart often result in unintended muscle tension, slack, or even damage, particularly when prescribed motion paths are derived directly from these imperfect models.

Core Contributions

1. Development of AIC: By implementing AIC, inspired by the human body's reciprocal innervation, the researchers devised a method to dynamically adjust muscle stiffness based on whether a muscle acts as an agonist or antagonist in a given motion. This technique harnesses muscle jacobian matrices to discern and appropriately scale muscle stiffness, reducing unnecessary antagonistic muscle tension and preventing slack.
2. Experimental Verification: The effectiveness of AIC was examined through a series of experiments on Kengoro, a tendon-driven musculoskeletal humanoid. The paper includes tests involving simple and complex joint movements, such as those in the shoulder and the scapula, and extends to dynamic tasks like pull-ups and prolonged dangling. The experiments showcased AIC's capacity to reduce muscle tension from notable peaks observed under traditional controls.
3. Comparison to Traditional Controls: Traditional controls, such as Muscle Stiffness Control (MSC), often presume accurate geometric models, failing when substantial model error is present. AIC avoids the pitfalls of model error by avoiding the assumption of geometric model accuracy, instead focusing solely on functional muscle relationships.

Implications and Future Directions

The implications of AIC extend beyond ensuring mechanical safety and robustness in humanoid robotics. This method represents a step toward more biologically plausible and efficient robotic systems, emulating human neuromechanical strategies for adaptable, damage-averse mobility. Given its focus on mitigating internal tension and maintaining mechanical integrity during wide range movements, AIC enables humanoid robots to perform complex tasks safely over extended periods.

Looking forward, integrating AIC with machine learning approaches to further model error adaptation could enhance the speed and precision of wide range motions. Such an integration might facilitate humanoids' ability to autonomously refine their internal models, achieving a deeper understanding and representation of their bodily mechanics, akin to proprioceptive learning in humans.

Conclusion

The development of AIC based on human-like reciprocal inhibition offers substantial improvements in the motion capabilities of tendon-driven musculoskeletal humanoids. This method models a significant advance over conventional control systems, enabling safer and more reliable humanoid operations across demanding kinetic tasks. Consequently, this work represents an important contribution to advancing humanoid robotics towards more human-like and versatile capabilities.