Calibration of a physics-based model of an anthropomimetic robot using Evolution Strategies

Abstract

Abstract-The control of tendon-driven and, in particular, of anthropomimetic robots using techniques from traditional robotics remains a very challenging task [1,2]. Hence, we previously proposed to employ physics-based simulation engines to simulate the complex dynamics of this emerging class of robots [3] and to use the simulation model as an internal model for robot control [4]. This approach, however, relies on an accurate model to be successful.In this paper, we present the automated, steady-state pose calibration of a physics-based, anthropomimetic robot model using a (µ, λ)-Evolution Strategy. For the acquisition of the poses of the physical robot, a stereo-vision, infrared-marker based motion capture system with real-time capabilities was developed. The employed (µ, λ)-Evolution Strategy uses a Gaussian-based, non-isotropic, self-adapting mutation operator to explore the search space and reduce the simulation-reality gap. The obtained results are impressive, resulting in a reduction of joint angle errors in the range of one to two orders of magnitude and an absolute joint angle error of 0.5• -4.5• per pose evaluated.

Adaptive neural network Dynamic Surface Control: An evaluation on the musculoskeletal robot Anthrob

“…To remain comparable throughout the experiments, high-level control was executed at a frequency of 200 Hz and low-level control at 1 kHz. Joint angles were measured by an analog potentiometer in the elbow joint and a stereo camera system with infrared markers on the fixed scapula and the humerus to obtain the state of the spherical shoulder joint [23]. Even though this motion capture system featured comparably low latencies of 4.3 ms in average, the frame rate was limited to 60 fps which complicated the numerical differentiation of the joint angle to obtain joint velocities.…”

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

“…We hypothesized that this is not due to the physics-based simulation approach per se, but rather due to: (i) the hand-crafted nature of the robot, which complicated the derivation of an accurate skeleton model, and (ii) the modeling simplifications that were necessary to simulate the complex dynamics of the tendon-driven muscles (as outlined in the previous sections). To evaluate this conjecture, we developed an automated, steadystate pose calibration algorithm based on a (A, E) evolution strategy (ES) [61]. For the acquisition of the poses of the physical robot, a stereo-vision, infrared-marker-based motion capture system with real-time capabilities was developed, and a Gaussian-based, non-isotropic, self-adapting mutation operator was used by the ES to explore the search space [6].…”

Anthrob - A printed anthropomimetic robot

“…Therefore, we developed a high-speed, stereo-vision motion capture system that uses infrared light and retroreflective markers, mounted to the scapula and humerus of the robot (not shown in Fig. 1) to track the joint position and velocity of the shoulder joint (see [13]). In human, the density of discriminative receptors is the "greatest on the hairless (glabrous) skin on the fingers, the palmar surface of the hand, the sole of the foot and the lips" [11].…”