Musculoskeletal AutoEncoder: A Unified Online Acquisition Method of Intersensory Networks for State Estimation, Control, and Simulation of Musculoskeletal Humanoids

Abstract

While the musculoskeletal humanoid has various biomimetic benefits, the modeling of its complex structure is difficult, and many learning-based systems have been developed so far. There are various methods, such as control methods using acquired relationships between joints and muscles represented by a data table or neural network, and state estimation methods using Extended Kalman Filter or table search. In this study, we construct a Musculoskeletal AutoEncoder representing the relationship among joint angles, muscle tensions, and muscle lengths, and propose a unified method of state estimation, control, and simulation of musculoskeletal humanoids using it. By updating the Musculoskeletal AutoEncoder online using the actual robot sensor information, we can continuously conduct more accurate state estimation, control, and simulation than before the online learning. We conducted several experiments using the musculoskeletal humanoid Musashi, and verified the effectiveness of this study.

Realization of Seated Walk by a Musculoskeletal Humanoid with Buttock-Contact Sensors From Human Constrained Teaching

“…Since the muscle wires are made of Dyneema ® , an abrasion resistant synthetic fiber, and are surrounded by a soft foam cover, their elasticity provides the flexibility of the body. By learning the relationship between muscle length, muscle tension, and joint angle, it is possible to control the joint angle [20]- [22]. However, due to the effects of friction and hysteresis, it is not always possible to control the joint angle accurately enough.…”

Hardware Design and Learning-Based Software Architecture of Musculoskeletal Wheeled Robot Musashi-W for Real-World Applications

“…Static body schema learning [12] is a learning mechanism for motion control of musculoskeletal bodies. By learning the relationship between joint angle θ, muscle tension f , and muscle length l, the robot is able to calculate the muscle length to achieve the desired joint angle and muscle tension, and to estimate unobservable joint angles.…”

“…One of the reasons for this is the difficulty of bipedal walking in a flexible body. The flexible and complex bodies of musculoskeletal humanoids are difficult to control, and while various learning-based control methods have been proposed [11], [12], none have yet succeeded in walking control. Therefore, real-world applications that take advantage of the muscle redundancy, variable stiffness control, and various biomimetic features of the musculoskeletal structure have not been conducted.…”

“…In addition, having a mechanum-wheeled base and a linear motion mechanism enables a wide range of stable movements to take advantage of the features of the musculoskeletal upper limbs. As for the software, in order to cope with the flexible and complicated hardware configurations, we constructed a learning system based on body schema learning [12], [13]. Static body schema learning [12] is used for basic motion controls of the musculoskeletal body, and dynamic body schema learning [13] is applied for more dynamic motions involving tools and target objects.…”

“…As for the software, in order to cope with the flexible and complicated hardware configurations, we constructed a learning system based on body schema learning [12], [13]. Static body schema learning [12] is used for basic motion controls of the musculoskeletal body, and dynamic body schema learning [13] is applied for more dynamic motions involving tools and target objects. In addition, we incorporate reflex controls to prevent the increase of internal force between antagonist muscles when performing a long series of motions [14], [15].…”