Fig 6 - uploaded by Kaiyu Deng
Content may be subject to copyright.
Joint motion for non-resetting test. The stimulus was applied from 2s to 2.1s at left hip pattern formation extensor neuron.
Source publication
We present a synthetic nervous system modeling mammalian locomotion using separate central pattern generator and pattern formation layers. The central pattern generator defines the rhythm of locomotion and the timing of extensor and flexor phase. We also investigated the capability of the pattern formation network to operate using muscle synergies...
Contexts in source publication
Context 1
... applied −10 nA tonic stimulus to inhibit the left hip pattern formation extensor neuron from 2s to 2.1s. Figure 6 presents the joint motion corresponding to this stimulus. The rat hind leg produced hopping motions until 2s, before the inhibitory stimulus was applied. ...
Context 2
... level. The left hip restarts extension after the stimulus ends, and generates a fast step to compensate for the delay. There is a small phase shift after the stimulus ends, but a short time later, the motion rhythm quickly returned to the original motion rhythm. This brief phase shifting is caused by feedback coordinating the joint as shown in Fig. 6, the knee and the ankle joint also shift their phase accordingly to produce steady ...
Context 3
... by the stimulus. So after the stimulus ends, the extension phase is still ongoing. The pattern formation extensor neuron, driven by the rhythm generator, undergoes a swift hyper-polarization and produces more neuron activation than normal, compensating for the delay, which leads to an overshoot of exten- sion in the hip joint as observed on Fig. 6. However, the rhythm generator is affected by afferent feedback for the following few cycles, which results in some phase shifting. However, the influence of the feedback is reduced by the intrinsic rhythmogenic proper- ties of the rhythm generator, so the step cycle rhythm recovers. ...
Context 4
... was the establishment of a simple muscle synergy, coordi- nating the ankle and knee motion by using the same pattern formation layer for both joints. This is an important first step towards synergy-based control, which is an increas- ingly common idea in motor control [15,16]. Our synthetic nervous system successfully produces steady hopping (Fig. 6) and alternating stepping (Fig. 8) and the hip and knee joint motion patterns for walking match the animal data. However, the ankle does not. This is possibly the result of simplifications we made in our biomechanical model. However, the advantage of these simplifications is much greater simulation speed which enables more rapid tuning ...
Context 5
... applied −10 nA tonic stimulus to inhibit the left hip pattern formation extensor neuron from 2s to 2.1s. Figure 6 presents the joint motion corresponding to this stimulus. The rat hind leg produced hopping motions until 2s, before the inhibitory stimulus was applied. At that moment, the hip joint was still in its flexion phase, but a short time after that while the right hip joint starts extension, the stimulus prolonged the flexion phase, causing the left leg to lag behind the master phase of the pattern generation level. The left hip restarts extension after the stimulus ends, and generates a fast step to compensate for the delay. There is a small phase shift after the stimulus ends, but a short time later, the motion rhythm quickly returned to the original motion rhythm. This brief phase shifting is caused by feedback coordinating the joint as shown in Fig. 6, the knee and the ankle joint also shift their phase accordingly to produce steady ...
Context 6
... applied −10 nA tonic stimulus to inhibit the left hip pattern formation extensor neuron from 2s to 2.1s. Figure 6 presents the joint motion corresponding to this stimulus. The rat hind leg produced hopping motions until 2s, before the inhibitory stimulus was applied. At that moment, the hip joint was still in its flexion phase, but a short time after that while the right hip joint starts extension, the stimulus prolonged the flexion phase, causing the left leg to lag behind the master phase of the pattern generation level. The left hip restarts extension after the stimulus ends, and generates a fast step to compensate for the delay. There is a small phase shift after the stimulus ends, but a short time later, the motion rhythm quickly returned to the original motion rhythm. This brief phase shifting is caused by feedback coordinating the joint as shown in Fig. 6, the knee and the ankle joint also shift their phase accordingly to produce steady ...
Context 7
... better understand this rhythm behavior, we inspected the neuron activities of the left hind leg. As Fig. 7 depicts, the extensor neuron in the pattern formation network is inhibited by the stimulus, reducing the activation time of the extensor motoneuron. The cycle timing generated by the rhythm generator is not influenced by the stimulus. So after the stimulus ends, the extension phase is still ongoing. The pattern formation extensor neuron, driven by the rhythm generator, undergoes a swift hyper-polarization and produces more neuron activation than normal, compensating for the delay, which leads to an overshoot of exten- sion in the hip joint as observed on Fig. 6. However, the rhythm generator is affected by afferent feedback for the following few cycles, which results in some phase shifting. However, the influence of the feedback is reduced by the intrinsic rhythmogenic proper- ties of the rhythm generator, so the step cycle rhythm recovers. ...
Context 8
... this work, we present improvements to our previous neural model of rat locomotion [5]. We made two primary improvements. First, we implemented a two-layer CPG structure, which enables our network to exhibit both phase-nonresetting and phase- resetting deletions. This is an important improvement for two reasons. First, our model is more biologically accurate in that it captures more phenomena observed in the animal. Second, this new model can more robustly control its stepping phase, which would be advantageous for a legged robot controlled by such a system, such as in Hunt et al. [6]. The second improvement was the establishment of a simple muscle synergy, coordi- nating the ankle and knee motion by using the same pattern formation layer for both joints. This is an important first step towards synergy-based control, which is an increas- ingly common idea in motor control [15,16]. Our synthetic nervous system successfully produces steady hopping (Fig. 6) and alternating stepping (Fig. 8) and the hip and knee joint motion patterns for walking match the animal data. However, the ankle does not. This is possibly the result of simplifications we made in our biomechanical model. However, the advantage of these simplifications is much greater simulation speed which enables more rapid tuning of parameters in the neural network using our Matlab toolbox SIMSCAN [18]. More detailed modeling will be needed to match the ankle ...
Similar publications
We aim to design a neuromorphic controller for the locomotion of a quadruped robot with muscle-driven leg mechanisms. To this end, we use a simulated cat model; each leg of the model is equipped with three joints driven by six muscle models incorporating two-joint muscles. For each leg, we use a two-level central pattern generator consisting of a r...
Citations
... In Deng et al. [68], an SNS was designed to control a biomechanical simulation of rat hindlimbs, with the network and body dynamics simulated using AnimatLab [37]. Here we reimplement this SNS using SNS-Toolbox and interface it with a new biomechanical model implemented in the physics simulator Mujoco [49]. ...
... In Deng et al. [68], the rat hindlimbs were modeled in AnimatLab [37] using simplified box geometry and a pair of linear Hill muscles for each joint. We replicate this model in Mujoco [49] using a three-dimensional model of bone geometry [72] and non-linear Hill muscles (shown in Figure 6C). ...
... On each step, muscle tensions from Mujoco are first formatted as Ia and Ib feedback for the SNS, then the outputs of the SNS are mapped via Equation (37) into muscle activations for the Mujoco model. The network parameters are exactly the same in this work as in Deng et al. [68], and result in oscillatory motor behavior (the original joint angles from AnimatLab can be seen in Figure A2). The overall leg oscillation occurs at half the frequency of the original model, and the joints exhibit scaled and shifted trajectories. ...
One developing approach for robotic control is the use of networks of dynamic neurons connected with conductance-based synapses, also known as Synthetic Nervous Systems (SNS). These networks are often developed using cyclic topologies and heterogeneous mixtures of spiking and non-spiking neurons, which is a difficult proposition for existing neural simulation software. Most solutions apply to either one of two extremes, the detailed multi-compartment neural models in small networks, and the large-scale networks of greatly simplified neural models. In this work, we present our open-source Python package SNS-Toolbox, which is capable of simulating hundreds to thousands of spiking and non-spiking neurons in real-time or faster on consumer-grade computer hardware. We describe the neural and synaptic models supported by SNS-Toolbox, and provide performance on multiple software and hardware backends, including GPUs and embedded computing platforms. We also showcase two examples using the software, one for controlling a simulated limb with muscles in the physics simulator Mujoco, and another for a mobile robot using ROS. We hope that the availability of this software will reduce the barrier to entry when designing SNS networks, and will increase the prevalence of SNS networks in the field of robotic control.
The current in vitro experiments reproducing food oral processing are not fully representative of the oral anatomy and of the biomechanics of the human tongue. Creating an in vitro device imitating the intricate interfacial and mechanical features, as well as the combination of motions of the human tongue, is critical to gain a quantitative understanding of how fluids and foods interact with oral surfaces. We present here a unique pneumatic three-dimensional multi-degree-of-freedom soft robotic actuator – the Biomimetic Soft Robotic Tongue (BSRT) – that embodies the mechanical performance and wettability of the human tongue, with realistic movements, and the surface roughness of the human tongue papillae. The present study presents a promising technology that can be used to realistically simulate the bolus dynamics in a variety of individual conditions, paving the way to design personalized food products meeting individual needs.KeywordsBiomimetic tonguesoft roboticsoral processingoral propulsionsemi-solid foods
This paper presents a biologically inspired control system developed for maintaining balance in a simulated human atop an oscillating platform. This work advances our previous research by adapting a human balance controller to an inverted pendulum and controlled by linear-Hill muscle models. To expedite neuron/synapse parameter value selection, we employ a novel two-stage process that pairs a previously developed analytic method with particle swarm optimization. Using the parameter values found analytically as inputs for particle swarm optimization (PSO), we take advantage of the benefits of each method while avoiding their pitfalls. Our results show that PSO optimization allowed improved balance control from modest (<10%) changes to the synaptic parameters. The improved performance was accompanied by muscle coactivations, however, and further refinement is needed to better align overall behavior of the neural controller with biological systems.KeywordsNeural ControllerBalance ControlFunctional Subnetwork ApproachParticle Swarm Optimization
Current initialization methods for artificial neural networks (ANNs) assume full connectivity between network layers. We propose that a bio-inspired initialization method for establishing connections between neurons in an artificial neural network will produce more accurate results relative to a fully connected network. We demonstrate four implementations of a novel, stochastic method for generating sparse connections in spatial, growth-based connectivity (GBC) maps. Connections in GBC maps are used to generate initial weights for neural networks in a deep learning compatible framework. These networks, designated as Growth-Initialized Neural Networks (GrINNs), have sparse connections between the input layer and the hidden layer. GrINNs were tested with user-specified nominal connectivity percentages ranging from 5–45%, resulting in unique connectivity percentages ranging from 4–28%. For reference, fully connected networks are defined as having 100% unique connectivity within this context. GrINNs with nominal connectivity percentages \(\ge \)20% produced better accuracy than fully connected ANNs when trained and tested on the MNIST dataset.
