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

(A) Neural mechanisms implemented on the bio-inspired hexapod robot AMOSII. The yellow circle (CPG) represents the neural locomotion control mechanism (see Supplementary Figure 1). The gray circles (RF1, 2, 3, …, 6) represent the reservoir-based adaptive forward models. The green circles (SE1, 2, 3, …, 6) represent searching and elevation control modules. The orange circles represent leg joints where TRi, CRi, FRi are TC-, CTr- and FTi-joints of the right front leg (i = 1), right middle leg (i = 2), right hind leg (i = 3) and TLi, CLi, FLi are left front leg (i = 1), left middle leg (i = 2), left hind leg (i = 3), respectively. BJ is a backbone joint. The orange arrow lines indicate the motor signals which are converted to joint angles for controlling motor positions. The black arrow lines indicate error signals. The green arrow lines indicate signals for adapting joint movements to deal with different circumstances. (B) An example of the reservoir-based adaptive forward model. The dashed frame shows a zoomed in view of a single reservoir neuron. In this setup, the input to each of the reservoir network comes from the CTr-joint of the respective leg. The reservoir learns to produce the expected foot contact signal for three different walking gaits (z1, z2, z3). The signals of the output neurons are combined and compared to the actual foot contact sensory signal. The error from the comparison is transmitted to an integrator unit. The unit accumulates the error over time. The accumulated error is finally used to adapt joint movements through searching and elevation control.
© Copyright Policy
Related In: Results  -  Collection

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

Figure 2: (A) Neural mechanisms implemented on the bio-inspired hexapod robot AMOSII. The yellow circle (CPG) represents the neural locomotion control mechanism (see Supplementary Figure 1). The gray circles (RF1, 2, 3, …, 6) represent the reservoir-based adaptive forward models. The green circles (SE1, 2, 3, …, 6) represent searching and elevation control modules. The orange circles represent leg joints where TRi, CRi, FRi are TC-, CTr- and FTi-joints of the right front leg (i = 1), right middle leg (i = 2), right hind leg (i = 3) and TLi, CLi, FLi are left front leg (i = 1), left middle leg (i = 2), left hind leg (i = 3), respectively. BJ is a backbone joint. The orange arrow lines indicate the motor signals which are converted to joint angles for controlling motor positions. The black arrow lines indicate error signals. The green arrow lines indicate signals for adapting joint movements to deal with different circumstances. (B) An example of the reservoir-based adaptive forward model. The dashed frame shows a zoomed in view of a single reservoir neuron. In this setup, the input to each of the reservoir network comes from the CTr-joint of the respective leg. The reservoir learns to produce the expected foot contact signal for three different walking gaits (z1, z2, z3). The signals of the output neurons are combined and compared to the actual foot contact sensory signal. The error from the comparison is transmitted to an integrator unit. The unit accumulates the error over time. The accumulated error is finally used to adapt joint movements through searching and elevation control.

Mentions: We design, six identical adaptive RNN-based forward models (RF1, 2, 3, …, 6), one for each leg of the walking robot (Figure 2A). These serve the purpose of online sensorimotor prediction as well as state estimation. Specifically, each forward model learns to correctly transform the efference copy of the actual motor signal for each leg joint (i.e., here the CTr-joint motor signal)1, into an expected or predicted sensory signal. This predicted signal is then compared with the actual incoming sensory feedback signals (i.e., here the foot contact signal—Figure 2B, of each leg) and, based on the error accumulated over time, it triggers the appropriate action (searching or elevation) and modulate the locomotive behavior of the robot. Each forward model is based on a random RNN architecture of the self-adaptive reservoir network type (Dasgupta et al., 2013; Dasgupta, 2015). Due to the presence of rich recurrent feedback connections, the dynamic reservoir and intrinsic homeostatic adaptations, the network exhibits a wide repertoire of non-linear activity and long fading memory. This can be primarily exploited for the purpose of specific leg joint-motor signal transformation, act as motor memory and for the prediction of sensorimotor patterns arising in the current context.


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)

(A) Neural mechanisms implemented on the bio-inspired hexapod robot AMOSII. The yellow circle (CPG) represents the neural locomotion control mechanism (see Supplementary Figure 1). The gray circles (RF1, 2, 3, …, 6) represent the reservoir-based adaptive forward models. The green circles (SE1, 2, 3, …, 6) represent searching and elevation control modules. The orange circles represent leg joints where TRi, CRi, FRi are TC-, CTr- and FTi-joints of the right front leg (i = 1), right middle leg (i = 2), right hind leg (i = 3) and TLi, CLi, FLi are left front leg (i = 1), left middle leg (i = 2), left hind leg (i = 3), respectively. BJ is a backbone joint. The orange arrow lines indicate the motor signals which are converted to joint angles for controlling motor positions. The black arrow lines indicate error signals. The green arrow lines indicate signals for adapting joint movements to deal with different circumstances. (B) An example of the reservoir-based adaptive forward model. The dashed frame shows a zoomed in view of a single reservoir neuron. In this setup, the input to each of the reservoir network comes from the CTr-joint of the respective leg. The reservoir learns to produce the expected foot contact signal for three different walking gaits (z1, z2, z3). The signals of the output neurons are combined and compared to the actual foot contact sensory signal. The error from the comparison is transmitted to an integrator unit. The unit accumulates the error over time. The accumulated error is finally used to adapt joint movements through searching and elevation control.
© Copyright Policy
Related In: Results  -  Collection

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

Figure 2: (A) Neural mechanisms implemented on the bio-inspired hexapod robot AMOSII. The yellow circle (CPG) represents the neural locomotion control mechanism (see Supplementary Figure 1). The gray circles (RF1, 2, 3, …, 6) represent the reservoir-based adaptive forward models. The green circles (SE1, 2, 3, …, 6) represent searching and elevation control modules. The orange circles represent leg joints where TRi, CRi, FRi are TC-, CTr- and FTi-joints of the right front leg (i = 1), right middle leg (i = 2), right hind leg (i = 3) and TLi, CLi, FLi are left front leg (i = 1), left middle leg (i = 2), left hind leg (i = 3), respectively. BJ is a backbone joint. The orange arrow lines indicate the motor signals which are converted to joint angles for controlling motor positions. The black arrow lines indicate error signals. The green arrow lines indicate signals for adapting joint movements to deal with different circumstances. (B) An example of the reservoir-based adaptive forward model. The dashed frame shows a zoomed in view of a single reservoir neuron. In this setup, the input to each of the reservoir network comes from the CTr-joint of the respective leg. The reservoir learns to produce the expected foot contact signal for three different walking gaits (z1, z2, z3). The signals of the output neurons are combined and compared to the actual foot contact sensory signal. The error from the comparison is transmitted to an integrator unit. The unit accumulates the error over time. The accumulated error is finally used to adapt joint movements through searching and elevation control.
Mentions: We design, six identical adaptive RNN-based forward models (RF1, 2, 3, …, 6), one for each leg of the walking robot (Figure 2A). These serve the purpose of online sensorimotor prediction as well as state estimation. Specifically, each forward model learns to correctly transform the efference copy of the actual motor signal for each leg joint (i.e., here the CTr-joint motor signal)1, into an expected or predicted sensory signal. This predicted signal is then compared with the actual incoming sensory feedback signals (i.e., here the foot contact signal—Figure 2B, of each leg) and, based on the error accumulated over time, it triggers the appropriate action (searching or elevation) and modulate the locomotive behavior of the robot. Each forward model is based on a random RNN architecture of the self-adaptive reservoir network type (Dasgupta et al., 2013; Dasgupta, 2015). Due to the presence of rich recurrent feedback connections, the dynamic reservoir and intrinsic homeostatic adaptations, the network exhibits a wide repertoire of non-linear activity and long fading memory. This can be primarily exploited for the purpose of specific leg joint-motor signal transformation, act as motor memory and for the prediction of sensorimotor patterns arising in the current context.

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