Data-driven artificial and spiking neural networks for inverse kinematics in neurorobotics

doi:10.1016/j.patter.2021.100391

Patterns

Volume 3, Issue 1, 14 January 2022, 100391

https://doi.org/10.1016/j.patter.2021.100391 Get rights and content

Under a Creative Commons license

open access

Highlights

•
Inverse kinematics lies at the foundation of robot motion planning
•
Artificial and spiking neural networks were used for data-driven inverse kinematics
•
A data-driven approach allows robust performance in convoluted environments

The bigger picture

Although artificial intelligence has emerged as the focal point for countless state-of-the-art developments, in many ways, its performance is nullified when compared with biological intelligence, particularly in terms of energy efficiency, robustness, versatility, and adaptivity. Therefore, neuromorphic (brain-inspired) computing has been utilized in numerous applications, particularly in robotics. Here, we uniquely address one of the most fundamental challenges in robotics: inverse kinematics in convoluted environments. Inverse kinematics is an underdetermined computational process for deriving a robot's configuration given its desired target position in space. A brain-inspired efficient implementation of inverse kinematics is, therefore, an important stepping stone in neurorobotics. Underdetermined inverse problems are also fundamental in other fields, ranging from medical imaging to hydrology and pharmacokinetics.

Summary

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.

Graphical abstract

Data Science Maturity

DSML 2: Proof-of-Concept: Data science output has been formulated, implemented, and tested for one domain/problem

Keywords

spiking neural networks

neural engineering framework

robotic arm

online learning

underdetermined systems

redundancy resolution

artificial neural networks

neuromorphic engineering

Intel Loihi

NVIDIA Xavier

Data and code availability

Data and code is available at https://github.com/NBELab/Patterns_2021.

Cited by (0)

⁴: Lead contact