Reproducing Five Motor Behaviors in a Salamander Robot With Virtual Muscles and a Distributed CPG Controller Regulated by Drive Signals and Proprioceptive Feedback

Abstract

Diverse locomotor behaviors emerge from the interactions between the spinal central pattern generator (CPG), descending brain signals and sensory feedback. Salamander motor behaviors include swimming, struggling, forward underwater stepping, and forward and backward terrestrial stepping. Electromyographic and kinematic recordings of the trunk show that each of these five behaviors is characterized by specific patterns of muscle activation and body curvature. Electrophysiological recordings in isolated spinal cords show even more diverse patterns of activity. Using numerical modeling and robotics, we explored the mechanisms through which descending brain signals and proprioceptive feedback could take advantage of the flexibility of the spinal CPG to generate different motor patterns. Adapting a previous CPG model based on abstract oscillators, we propose a model that reproduces the features of spinal cord recordings: the diversity of motor patterns, the correlation between phase lags and cycle frequencies, and the spontaneous switches between slow and fast rhythms. The five salamander behaviors were reproduced by connecting the CPG model to a mechanical simulation of the salamander with virtual muscles and local proprioceptive feedback. The main results were validated on a robot. A distributed controller was used to obtain the fast control loops necessary for implementing the virtual muscles. The distributed control is demonstrated in an experiment where the robot splits into multiple functional parts. The five salamander behaviors were emulated by regulating the CPG with two descending drives. Reproducing the kinematics of backward stepping and struggling however required stronger muscle contractions. The passive oscillations observed in the salamander's tail during forward underwater stepping could be reproduced using a third descending drive of zero to the tail oscillators. This reduced the drag on the body in our hydrodynamic simulation. We explored the effect of local proprioceptive feedback during swimming and forward terrestrial stepping. We found that feedback could replace or reduce the need for different drives in both cases. It also reduced the variability of intersegmental phase lags toward values appropriate for locomotion. Our work suggests that different motor behaviors do not require different CPG circuits: a single circuit can produce various behaviors when modulated by descending drive and sensory feedback.

Editorial: Biological and Robotic Inter-Limb Coordination

“…The salamander constitutes a model animal for focusing on the following two issues, representing an evolutionary process of moving from water to land: 1) versatile behavior generation against a changing environment, based on CPGs coordinated by both descending signals and sensory feedback; and 2) body-limb coordination, i.e., coordination between undulatory movements of the body and leg movements based on the salamander’s characteristic morphology. Using a robot, Knüsel et al (2020) were able to reproduce the five motor behaviors observed in salamanders: swimming, struggling, forward underwater stepping, and forward and backward terrestrial stepping. A mathematical model is presented that allows the robot to switch between various motor patterns using a neural circuit with descending brain signals and proprioceptive feedback as input.…”

Course Control of a Manta Robot Based on Amplitude and Phase Differences

“…By adjusting the two input signals to the CPG model, the speed, direction and gait type of the robot can be adjusted, smooth switching from a crawling gait to a swimming gait can be completed, and a turning action can also be completed. Based on the Salamander I robot, a Salamander II robot was established [38]. Using a single CPG circuit, it realized five behaviors: swimming, struggling, forward underwater walking, forward and backward ground walking, and integrated ontology feedback to optimize swimming speed.…”

Effects of external signals on neural oscillator stability

“…Liu et al, (2012) and Liu et al, (2008) used a CPG to generate bipedal robot locomotion. Also, Habu et al, (2019) used a CPG for a quadruped robot, Knüsel et al, (2020) did for a multi-segmental salamander robot, and Avrin et al, (2017) developed a visual control model of rhythmic ball bouncing using CPG. CPG allows us not only to make a signal with a constant period, but also to change period smoothly with a method of muscle synergy (Aoi et al, 2019), two-layered system (Saputra et al, 2019), and triple-layered system (Qiao et al, 2017).…”