Echo State Networks for Estimating Exteroceptive Conditions From Proprioceptive States in Quadruped Robots

Abstract

We propose a methodology based on reservoir computing for mapping local proprioceptive information acquired at the level of the leg joints of a simulated quadruped robot into exteroceptive and global information, including both the ground reaction forces at the level of the different legs and information about the type of terrain traversed by the robot. Both dynamic estimation and terrain classification can be achieved concurrently with the same reservoir computing structure, which serves as a soft sensor device. Simulation results are presented together with preliminary experiments on a real quadruped robot. They demonstrate the suitability of the proposed approach for various terrains and sensory system fault conditions. The strategy, which belongs to the class of data-driven models, is independent of the robotic mechanical design and can easily be generalized to different robotic structures.

Gait Transition from Pacing by a Quadrupedal Simulated Model and Robot with Phase Modulation by Vestibular Feedback

“…By combining the proposed method of the autonomous leg phase modulation with the methods to determine the motion plan by machine learning, more advanced locomotions such as turning and climbing are expected. Furthermore, a method has been proposed to achieve walking by learning in a simulation and implementing the result on a robot [51,52]. This method reduces the number of sensors and does not require many experiments, but it does not allow it to run through unperceived obstacles.…”

Intelligent design for quadruped robot on a dynamic flexible surface with an active spine

“…This decomposition results in two bipedal systems, into which Reservoir Computing is integrated to model the dynamics of the quadrupedal robot via spine state measurement. 9 Once the dynamic behavior of the quadruped robot with spine is known, the Lyapunov Control algorithm is employed to ensure guaranteed stability. 10,11 Additionally, a feedback loop is utilized to verify whether the robot center of mass is mapping onto the dynamic surface accurately 8 .…”

“…1,13 The bending sensors' signals are then processed through a Reservoir Computing layer that employs Echo State Networks (ESN) to generate a time series prediction of the system's state. 9,14 The Reservoir Computing layer comprises three major components: the input, reservoir layers, and readout. The input layer maps the bending sensor signals into the Reservoir Computing space, while the reservoir layer is composed of a randomly connected network of neurons that transforms the input into a high-dimensional feature space.…”

“…The input layer maps the bending sensor signals into the Reservoir Computing space, while the reservoir layer is composed of a randomly connected network of neurons that transforms the input into a high-dimensional feature space. 14,15 The Reservoir Computing layer then feeds the transformed input into a linear readout layer to estimate the system's response, the benefit for Reservoir Computing is that it able to analysis chaotic system which suitable for soft robotics dynamic modeling. 13,16 The only trainable part of reservoir computing is the readout, which is ridge node, can be trained by state of training and y[t], where the state of training is the activation of the reservoir trigger by x[t], 10 as shown in Figure 1.…”

Editorial: Biological and Robotic Inter-Limb Coordination

“…Robust robot state estimation and sensory event mistiming detection are important issues for adaptive inter-limb coordination. Accordingly, Calandra et al (2021) proposed a data-driven method using reservoir computing for translating local proprioceptive feedback, acquired at the leg joints of a simulated quadruped robot, into global exteroceptive information, which include both GRFs at the level of the different legs and information about the type of terrain traversed by the robot. This mechanism enables the robot to effectively estimate its walking state (i.e., estimating the GRFs from joint torques) and classify terrains for adaptive locomotion.…”