Computed muscle control for an anthropomimetic elbow joint

Abstract

Abstract-The soft robotics approach is widely considered to enable human-friendly robots which are able to work in our future homes and factories. Furthermore, achieving the smooth and natural movements of humans has become a hot topic in robotics, especially when robots are supposed to work in close proximity to humans. The anthropomimetic principle aims at mimicking not only the outside but also the inner mechanisms of the human body in humanoid robots. However, for this class of robots there exist as yet no scalable controllers that might make it possible to control a full body, or even several joints. A very similar problem is ongoing research in biomechanics which is the computation of muscle excitation patterns for coordinated movements. For this purpose, biomechanicists have developed computed muscle control which has proven a very scalable technique.In this paper, we demonstrate the adaptation of computed muscle control for a tendon-driven robot, comparing different methods for obtaining the muscle kinematics, as well as different low-level controllers. Results are shown for the implementation on a distributed control architecture and a single revolute elbow joint.

Anthrob - A printed anthropomimetic robot

“…Hence, a sensorized robotic platform was created that is capable of serving as a test bed for both neuroscientists and control engineers. This has already been demonstrated by the previously presented computed muscle control algorithm which was successfully evaluated on the robot [15]. However, we also hope that the application of this robot will not be limited to robotic applications per se but that it will ultimately help to shed new light on our understanding of human motion control.…”

“…To achieve a high degree of robustness during robot operation, each ECU To facilitate the development of high-level joint-or operational space controllers, a set of four low-level muscle controllers was developed and implemented on the ECUs. These are (i) a voltage control mode, (ii) a proportional-derivative (PD) motor position control mode, (iii) a proportionalintegral-derivative (PID) current control mode and (iv) a state-space muscle force controller, which has proven to be superior to conventional PD approaches [15]. All controllers are executed with a control frequency of 1 kHz, which is sufficient as the SEEs slow down the muscle dynamics.…”

Robotics library: An object-oriented approach to robot applications

“…In combination with the muscle Jacobian, [23] shows an application of this part of RL for tendon-driven robots.…”

Toward Anthropomimetic Robotics: Development, Simulation, and Control of a Musculoskeletal Torso

“…Hence, particular emphasis had to be put on the control laws to achieve stable and responsive control, especially for the tendon force control mode. Here, a state space controller was implemented that proved to be superior to conventional proportional-derivative (PD) approaches [30]. It can be designed directly in the discrete space domain, taking the control frequency into account at design time, utilizing the state space actuator model (see Section 4.2), by moving the closed loop poles to the desired values (Ackermannʼs formula [1]).…”