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

Average time to successfully overcome uneven terrains of different elasticity (hard, moderate, highly elastic). (A) Average success time for reservoir-based forward model. (B) Average success time for adaptive neuron forward model from Manoonpong et al. (2013). Here the whiskers indicate one standard deviation above and below the mean value. Note the difference in scale of the y-axis in both plots. The experimental surface here consisted of the rough terrain as presented in Figure 6 consisting of irregular undulations, however with varying degree of elasticity for the three cases.
© Copyright Policy
Related In: Results  -  Collection

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

Figure 10: Average time to successfully overcome uneven terrains of different elasticity (hard, moderate, highly elastic). (A) Average success time for reservoir-based forward model. (B) Average success time for adaptive neuron forward model from Manoonpong et al. (2013). Here the whiskers indicate one standard deviation above and below the mean value. Note the difference in scale of the y-axis in both plots. The experimental surface here consisted of the rough terrain as presented in Figure 6 consisting of irregular undulations, however with varying degree of elasticity for the three cases.

Mentions: In order to evaluate the performance of our adaptive reservoir forward model in comparison to the state of the art model recently presented in Manoonpong et al. (2013) (single recurrent neural with low-pass filter), we carried out simulation experiments with AMOSII walking on different types of surfaces. Specifically, after training on a flat surface (under normal conditions) we carried out 10 trials each with the robot walking on uneven terrains (laid with multiple obstacles of height 8 cm), having three different elastic properties4. The surfaces were divided into hard (1.0), moderately elastic (5.0) and highly elastic (10.0). A tetrapod walking gait was used in all three cases. Starting from a fixed position, we noted the total time taken by the robot to successfully cross the uneven terrain region and move into a flat surface region. As observed in Figures 10A,B, the reservoir forward model enables the robot to traverse the uneven region considerably faster as compared to the adaptive neuron forward model, in all three scenarios. Both the models can be seen to overcome the hard surface much better as compared to the elastic ones. This was expected due to the changes in surface stiffness resulting in additional forces on the robot legs. However, the reservoir model performance was considerably more robust with a mean difference in success time of 1.86 min for the hardest surface and approximately 2 min for the most elastic surface, cases. Given that the walking gait was fixed, here the success time can be thought as an indicator of the robot's energy efficiency. In the absence of additional body mechanisms to deal with changing surface stiffness, the reservoir based model outperforms the previous implementations of adaptive forward models by ≈25% on average. In the climbing and gap crossing scenarios, the performance of the two forward models are comparable (not shown here explicitly) unless there are significant changes in the ground reaction forces (e.g., climbing or crossing gaps on different types of terrain). As such the reservoir forward model offers a more generalized architecture for adaptive locomotion. Furthermore, as demonstrated previously, this model is also capable of robustly coping with missing motor information and a high degree of sensory noise; making use of the SARN internal memory and multiple timescales (Dasgupta, 2015). This was very difficult to achieve with the previous simple single recurrent neuron forward models. Moreover, the previous study also required that a separate forward model be learned for every different walking gait. Thus, creating a scalability issue for real robot implementations. Here, however, a single SARN can be trained online to predict the foot contact signals for multiple different walking gaits (here we show three gaits, but it can be easily extended to many more patterns—see Supplementary Figure 2, for tripod gait example).


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)

Average time to successfully overcome uneven terrains of different elasticity (hard, moderate, highly elastic). (A) Average success time for reservoir-based forward model. (B) Average success time for adaptive neuron forward model from Manoonpong et al. (2013). Here the whiskers indicate one standard deviation above and below the mean value. Note the difference in scale of the y-axis in both plots. The experimental surface here consisted of the rough terrain as presented in Figure 6 consisting of irregular undulations, however with varying degree of elasticity for the three cases.
© Copyright Policy
Related In: Results  -  Collection

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

Figure 10: Average time to successfully overcome uneven terrains of different elasticity (hard, moderate, highly elastic). (A) Average success time for reservoir-based forward model. (B) Average success time for adaptive neuron forward model from Manoonpong et al. (2013). Here the whiskers indicate one standard deviation above and below the mean value. Note the difference in scale of the y-axis in both plots. The experimental surface here consisted of the rough terrain as presented in Figure 6 consisting of irregular undulations, however with varying degree of elasticity for the three cases.
Mentions: In order to evaluate the performance of our adaptive reservoir forward model in comparison to the state of the art model recently presented in Manoonpong et al. (2013) (single recurrent neural with low-pass filter), we carried out simulation experiments with AMOSII walking on different types of surfaces. Specifically, after training on a flat surface (under normal conditions) we carried out 10 trials each with the robot walking on uneven terrains (laid with multiple obstacles of height 8 cm), having three different elastic properties4. The surfaces were divided into hard (1.0), moderately elastic (5.0) and highly elastic (10.0). A tetrapod walking gait was used in all three cases. Starting from a fixed position, we noted the total time taken by the robot to successfully cross the uneven terrain region and move into a flat surface region. As observed in Figures 10A,B, the reservoir forward model enables the robot to traverse the uneven region considerably faster as compared to the adaptive neuron forward model, in all three scenarios. Both the models can be seen to overcome the hard surface much better as compared to the elastic ones. This was expected due to the changes in surface stiffness resulting in additional forces on the robot legs. However, the reservoir model performance was considerably more robust with a mean difference in success time of 1.86 min for the hardest surface and approximately 2 min for the most elastic surface, cases. Given that the walking gait was fixed, here the success time can be thought as an indicator of the robot's energy efficiency. In the absence of additional body mechanisms to deal with changing surface stiffness, the reservoir based model outperforms the previous implementations of adaptive forward models by ≈25% on average. In the climbing and gap crossing scenarios, the performance of the two forward models are comparable (not shown here explicitly) unless there are significant changes in the ground reaction forces (e.g., climbing or crossing gaps on different types of terrain). As such the reservoir forward model offers a more generalized architecture for adaptive locomotion. Furthermore, as demonstrated previously, this model is also capable of robustly coping with missing motor information and a high degree of sensory noise; making use of the SARN internal memory and multiple timescales (Dasgupta, 2015). This was very difficult to achieve with the previous simple single recurrent neuron forward models. Moreover, the previous study also required that a separate forward model be learned for every different walking gait. Thus, creating a scalability issue for real robot implementations. Here, however, a single SARN can be trained online to predict the foot contact signals for multiple different walking gaits (here we show three gaits, but it can be easily extended to many more patterns—see Supplementary Figure 2, for tripod gait example).

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