TY - JOUR
T1 - Data-driven artificial and spiking neural networks for inverse kinematics in neurorobotics
AU - Volinski, Alex
AU - Zaidel, Yuval
AU - Shalumov, Albert
AU - DeWolf, Travis
AU - Supic, Lazar
AU - Ezra Tsur, Elishai
N1 - Funding Information: The authors thank the Applied Brain Research (ABR) team for the support; Intel Labs for granting us access to their neuromorphic cloud and technical support; and Andreea Danielescu and Timothy Shea from Accenture Labs for their insightful comments. This research was funded by Accenture Labs as part of Intel's INRC (Intel Neuromorphic Research Community) initiative and by the Open University of Israel research grant. E.E.T. A.V. Y.Z. and A.S. conceptualized the study; E.E.T. A.V. Y.Z. A.S. and T.D. developed the methodology; E.E.T. A.V. Y.Z. A.S. T.D. and L.S. performed the formal analysis; E.E.T. wrote the manuscript; E.E.T. A.V. Y.Z. A.S. T.D. and L.S. reviewed and edited the manuscript; E.E.T. supervised the study and acquired funding. The authors declare no competing interests. Funding Information: The authors thank the Applied Brain Research (ABR) team for the support; Intel Labs for granting us access to their neuromorphic cloud and technical support; and Andreea Danielescu and Timothy Shea from Accenture Labs for their insightful comments. This research was funded by Accenture Labs as part of Intel's INRC (Intel Neuromorphic Research Community) initiative and by the Open University of Israel research grant. Publisher Copyright: © 2021 The Author(s) © 2021 The Author(s). © 2021 The Author(s). © 2021 The Author(s).
PY - 2022/1/14
Y1 - 2022/1/14
N2 - Inverse kinematics is fundamental for computational motion planning. It is used to derive an appropriate state in a robot's configuration space, given a target position in task space. In this work, we investigate the performance of fully connected and residual artificial neural networks as well as recurrent, learning-based, and deep spiking neural networks for conventional and geometrically constrained inverse kinematics. We show that while highly parameterized data-driven neural networks with tens to hundreds of thousands of parameters exhibit sub-ms inference time and sub-mm accuracy, learning-based spiking architectures can provide reasonably good results with merely a few thousand neurons. Moreover, we show that spiking neural networks can perform well in geometrically constrained task space, even when configured to an energy-conserved spiking rate, demonstrating their robustness. Neural networks were evaluated on NVIDIA's Xavier and Intel's neuromorphic Loihi chip.
AB - Inverse kinematics is fundamental for computational motion planning. It is used to derive an appropriate state in a robot's configuration space, given a target position in task space. In this work, we investigate the performance of fully connected and residual artificial neural networks as well as recurrent, learning-based, and deep spiking neural networks for conventional and geometrically constrained inverse kinematics. We show that while highly parameterized data-driven neural networks with tens to hundreds of thousands of parameters exhibit sub-ms inference time and sub-mm accuracy, learning-based spiking architectures can provide reasonably good results with merely a few thousand neurons. Moreover, we show that spiking neural networks can perform well in geometrically constrained task space, even when configured to an energy-conserved spiking rate, demonstrating their robustness. Neural networks were evaluated on NVIDIA's Xavier and Intel's neuromorphic Loihi chip.
KW - DSML 2: Proof-of-Concept: Data science output has been formulated, implemented, and tested for one domain/problem
KW - Intel Loihi
KW - NVIDIA Xavier
KW - artificial neural networks
KW - neural engineering framework
KW - neuromorphic engineering
KW - online learning
KW - redundancy resolution
KW - robotic arm
KW - spiking neural networks
KW - underdetermined systems
UR - https://www.scopus.com/pages/publications/85122645233
U2 - 10.1016/j.patter.2021.100391
DO - 10.1016/j.patter.2021.100391
M3 - Article
C2 - 35079712
SN - 2666-3899
VL - 3
SP - 100391
JO - Patterns
JF - Patterns
IS - 1
M1 - 100391
ER -