Limits...
Distributed recurrent neural forward models with synaptic adaptation and CPG-based control for complex behaviors of walking robots.

Dasgupta S, Goldschmidt D, Wörgötter F, Manoonpong P - Front Neurorobot (2015)

Bottom Line: Biological study has revealed that such complex behaviors are a result of a combination of biomechanics and neural mechanism thus representing the true nature of embodied interactions.Inspired by these findings, we present here, an artificial bio-inspired walking system which effectively combines biomechanics (in terms of the body and leg structures) with the underlying neural mechanisms.Furthermore, we demonstrate that the newly developed recurrent network based approach to online forward models outperforms the adaptive neuron forward models, which have hitherto been the state of the art, to model a subset of similar walking behaviors in walking robots.

View Article: PubMed Central - PubMed

Affiliation: Institute for Physics - Biophysics, George-August-University Göttingen, Germany ; Bernstein Center for Computational Neuroscience, George-August-University Göttingen, Germany ; Laboratory for Neural Computation and Adaptation, Riken Brain Science Institute Saitama, Japan.

ABSTRACT
Walking animals, like stick insects, cockroaches or ants, demonstrate a fascinating range of locomotive abilities and complex behaviors. The locomotive behaviors can consist of a variety of walking patterns along with adaptation that allow the animals to deal with changes in environmental conditions, like uneven terrains, gaps, obstacles etc. Biological study has revealed that such complex behaviors are a result of a combination of biomechanics and neural mechanism thus representing the true nature of embodied interactions. While the biomechanics helps maintain flexibility and sustain a variety of movements, the neural mechanisms generate movements while making appropriate predictions crucial for achieving adaptation. Such predictions or planning ahead can be achieved by way of internal models that are grounded in the overall behavior of the animal. Inspired by these findings, we present here, an artificial bio-inspired walking system which effectively combines biomechanics (in terms of the body and leg structures) with the underlying neural mechanisms. The neural mechanisms consist of (1) central pattern generator based control for generating basic rhythmic patterns and coordinated movements, (2) distributed (at each leg) recurrent neural network based adaptive forward models with efference copies as internal models for sensory predictions and instantaneous state estimations, and (3) searching and elevation control for adapting the movement of an individual leg to deal with different environmental conditions. Using simulations we show that this bio-inspired approach with adaptive internal models allows the walking robot to perform complex locomotive behaviors as observed in insects, including walking on undulated terrains, crossing large gaps, leg damage adaptations, as well as climbing over high obstacles. Furthermore, we demonstrate that the newly developed recurrent network based approach to online forward models outperforms the adaptive neuron forward models, which have hitherto been the state of the art, to model a subset of similar walking behaviors in walking robots.

No MeSH data available.


Related in: MedlinePlus

Reservoir-to-readout weight adaptation during online learning. (A) Changes of 30 weights projecting to the first readout neuron (z1) of the forward model of the right front leg (R1) while walking with a wave gait. During this period, weights projecting to the second (z2) and third (z3) output neurons remain unchanged (i.e., they are zero). (B) Changes of the weights to z2 while walking with a tetrapod gait. During this period, the weights to z3 still remain unchanged and the weights to z1 converge to around zero. (C) Changes of the weights to z3 while walking with a caterpillar gait. During this period, the weights to z1 and z2 converge to around zero. At the end of each gait, all weights are stored such that they will be used for locomotion in different environments. The gray areas represent transition phases from one gait to another gait and the yellow areas represent convergence. The gait diagrams are shown on the right. They are observed from the motor signals of the CTr-joints (Figure 5). White areas indicate ground contact or stance phase and blue areas refer to no ground contact during swing phase. As frequency increases, some legs step in pairs (dashed enclosures). Here convergence implies no significant change in the vector norm of the readout weights.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4585172&req=5

Figure 4: Reservoir-to-readout weight adaptation during online learning. (A) Changes of 30 weights projecting to the first readout neuron (z1) of the forward model of the right front leg (R1) while walking with a wave gait. During this period, weights projecting to the second (z2) and third (z3) output neurons remain unchanged (i.e., they are zero). (B) Changes of the weights to z2 while walking with a tetrapod gait. During this period, the weights to z3 still remain unchanged and the weights to z1 converge to around zero. (C) Changes of the weights to z3 while walking with a caterpillar gait. During this period, the weights to z1 and z2 converge to around zero. At the end of each gait, all weights are stored such that they will be used for locomotion in different environments. The gray areas represent transition phases from one gait to another gait and the yellow areas represent convergence. The gait diagrams are shown on the right. They are observed from the motor signals of the CTr-joints (Figure 5). White areas indicate ground contact or stance phase and blue areas refer to no ground contact during swing phase. As frequency increases, some legs step in pairs (dashed enclosures). Here convergence implies no significant change in the vector norm of the readout weights.

Mentions: Subsequent to the pre-training phase, normal training of the reservoir-to-readout weights Wout was carried out using the online RLS learning algorithm with the same process of making the robot walk on a flat, regular terrain and sequential switching between the three gait patterns every 2500 time steps. As such, at any given point in time only one of the readout neurons (specific to the walking gait) are active. In this manner, synaptic weights projecting from reservoir to the first readout neuron (z1) corresponding to the foot contact signal prediction for the wave gait, and synaptic weights projecting to the second (z2) and third (z3) readout neurons corresponding to the foot contact signal prediction of the tetrapod and caterpillar gaits, are learned, respectively. Within this experimental setup, as observed from Figures 4A–C the readout weights corresponding to each gait converges very quickly, in less than the trial period of 2500 time steps3. As a result, every time the CTr-motor signal changes due to walking gait transformations, the RF associated with each leg learns to predict the expected foot contact signal robustly. The training process was carried out only once under normal walking conditions. This was subsequently used as the baseline in order to compare with the actual foot contact signals (sensory feedback) while walking under the situations of crossing a gap, climbing, and negotiating uneven terrains.


Distributed recurrent neural forward models with synaptic adaptation and CPG-based control for complex behaviors of walking robots.

Dasgupta S, Goldschmidt D, Wörgötter F, Manoonpong P - Front Neurorobot (2015)

Reservoir-to-readout weight adaptation during online learning. (A) Changes of 30 weights projecting to the first readout neuron (z1) of the forward model of the right front leg (R1) while walking with a wave gait. During this period, weights projecting to the second (z2) and third (z3) output neurons remain unchanged (i.e., they are zero). (B) Changes of the weights to z2 while walking with a tetrapod gait. During this period, the weights to z3 still remain unchanged and the weights to z1 converge to around zero. (C) Changes of the weights to z3 while walking with a caterpillar gait. During this period, the weights to z1 and z2 converge to around zero. At the end of each gait, all weights are stored such that they will be used for locomotion in different environments. The gray areas represent transition phases from one gait to another gait and the yellow areas represent convergence. The gait diagrams are shown on the right. They are observed from the motor signals of the CTr-joints (Figure 5). White areas indicate ground contact or stance phase and blue areas refer to no ground contact during swing phase. As frequency increases, some legs step in pairs (dashed enclosures). Here convergence implies no significant change in the vector norm of the readout weights.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4585172&req=5

Figure 4: Reservoir-to-readout weight adaptation during online learning. (A) Changes of 30 weights projecting to the first readout neuron (z1) of the forward model of the right front leg (R1) while walking with a wave gait. During this period, weights projecting to the second (z2) and third (z3) output neurons remain unchanged (i.e., they are zero). (B) Changes of the weights to z2 while walking with a tetrapod gait. During this period, the weights to z3 still remain unchanged and the weights to z1 converge to around zero. (C) Changes of the weights to z3 while walking with a caterpillar gait. During this period, the weights to z1 and z2 converge to around zero. At the end of each gait, all weights are stored such that they will be used for locomotion in different environments. The gray areas represent transition phases from one gait to another gait and the yellow areas represent convergence. The gait diagrams are shown on the right. They are observed from the motor signals of the CTr-joints (Figure 5). White areas indicate ground contact or stance phase and blue areas refer to no ground contact during swing phase. As frequency increases, some legs step in pairs (dashed enclosures). Here convergence implies no significant change in the vector norm of the readout weights.
Mentions: Subsequent to the pre-training phase, normal training of the reservoir-to-readout weights Wout was carried out using the online RLS learning algorithm with the same process of making the robot walk on a flat, regular terrain and sequential switching between the three gait patterns every 2500 time steps. As such, at any given point in time only one of the readout neurons (specific to the walking gait) are active. In this manner, synaptic weights projecting from reservoir to the first readout neuron (z1) corresponding to the foot contact signal prediction for the wave gait, and synaptic weights projecting to the second (z2) and third (z3) readout neurons corresponding to the foot contact signal prediction of the tetrapod and caterpillar gaits, are learned, respectively. Within this experimental setup, as observed from Figures 4A–C the readout weights corresponding to each gait converges very quickly, in less than the trial period of 2500 time steps3. As a result, every time the CTr-motor signal changes due to walking gait transformations, the RF associated with each leg learns to predict the expected foot contact signal robustly. The training process was carried out only once under normal walking conditions. This was subsequently used as the baseline in order to compare with the actual foot contact signals (sensory feedback) while walking under the situations of crossing a gap, climbing, and negotiating uneven terrains.

Bottom Line: Biological study has revealed that such complex behaviors are a result of a combination of biomechanics and neural mechanism thus representing the true nature of embodied interactions.Inspired by these findings, we present here, an artificial bio-inspired walking system which effectively combines biomechanics (in terms of the body and leg structures) with the underlying neural mechanisms.Furthermore, we demonstrate that the newly developed recurrent network based approach to online forward models outperforms the adaptive neuron forward models, which have hitherto been the state of the art, to model a subset of similar walking behaviors in walking robots.

View Article: PubMed Central - PubMed

Affiliation: Institute for Physics - Biophysics, George-August-University Göttingen, Germany ; Bernstein Center for Computational Neuroscience, George-August-University Göttingen, Germany ; Laboratory for Neural Computation and Adaptation, Riken Brain Science Institute Saitama, Japan.

ABSTRACT
Walking animals, like stick insects, cockroaches or ants, demonstrate a fascinating range of locomotive abilities and complex behaviors. The locomotive behaviors can consist of a variety of walking patterns along with adaptation that allow the animals to deal with changes in environmental conditions, like uneven terrains, gaps, obstacles etc. Biological study has revealed that such complex behaviors are a result of a combination of biomechanics and neural mechanism thus representing the true nature of embodied interactions. While the biomechanics helps maintain flexibility and sustain a variety of movements, the neural mechanisms generate movements while making appropriate predictions crucial for achieving adaptation. Such predictions or planning ahead can be achieved by way of internal models that are grounded in the overall behavior of the animal. Inspired by these findings, we present here, an artificial bio-inspired walking system which effectively combines biomechanics (in terms of the body and leg structures) with the underlying neural mechanisms. The neural mechanisms consist of (1) central pattern generator based control for generating basic rhythmic patterns and coordinated movements, (2) distributed (at each leg) recurrent neural network based adaptive forward models with efference copies as internal models for sensory predictions and instantaneous state estimations, and (3) searching and elevation control for adapting the movement of an individual leg to deal with different environmental conditions. Using simulations we show that this bio-inspired approach with adaptive internal models allows the walking robot to perform complex locomotive behaviors as observed in insects, including walking on undulated terrains, crossing large gaps, leg damage adaptations, as well as climbing over high obstacles. Furthermore, we demonstrate that the newly developed recurrent network based approach to online forward models outperforms the adaptive neuron forward models, which have hitherto been the state of the art, to model a subset of similar walking behaviors in walking robots.

No MeSH data available.


Related in: MedlinePlus