Learning Inverse Kinematics using Neural Computational Primitives on Neuromorphic Hardware

Abstract

Current low latency neuromorphic processing systems, and future ones based on ultra-low power mixed-signal circuits in advanced technology nodes and memristive nano-scale devices, hold great potential for developing autonomous artificial agents. However, the variable nature and low precision of the underlying hardware substrate pose severe challenges for robust and reliable performance. To address these challenges, we adopt hardware-friendly processing strategies based on brain-inspired computational primitives, such as triplet spike-timing dependent plasticity, basal ganglia inspired disinhibition, and cooperative-competitive networks. We demonstrate this approach by presenting an example of robust real-time motor control using a hardware spiking neural network implemented on a mixed-signal neuromorphic processor, trained to learn the inverse kinematics of a 2-joint robotic arm. The final system is able to perform low-latency control robustly and reliably using noisy silicon neurons. The spiking neural network, trained to control two joints of the iCub robot arm simulator, performs a continuous target reaching task with 97.93% accuracy, 33.96 ms network latency, 102.1 ms system latency and with an estimated power consumption of 26.92 µW. This work provides insights into how specific computational primitives used by real neural systems can be applied to neuromorphic computing for solving real-world engineering tasks. It represents a milestone in the design of end-to-end spiking robotic control systems, relying on event-driven sensory encoding, neuromorphic processing, and spiking motor control.

Bioinspired smooth neuromorphic control for robotic arms

“…Overall, this work supports ongoing efforts to develop real-time and robust neuromorphic solutions for robot control exhibiting desired behaviors [19]. The proposed methods could be combined with emerging solutions for the inverse-kinematics calculation of high-dimensional robotic arms [67,76] to provide end-to-end neuromorphic control. Unlike other methods, here we aimed to derive computational principles from the knowledge of how the brain attains optimal motor behavior and introduce them to the design of neuromorphic robotic controllers, adding to the growing pool of non-cognitive neuromorphic applications [1].…”

A bio-inspired hardware implementation of an analog spike-based hippocampus memory model

“…The DYNAP-SE hardware platform contains the desired network implementation. In addition, a computer-inthe-loop setup is necessary to implement the triplet STDP learning algorithm in DYNAP-SE [52]. DYNAP-SE returns the information from the network in real time to the PC and, based on this information, it calculates the changes of weights in those synapses with the triplet STDP mechanism.…”

A neuromorphic electronic artist for robotic painting

“…To learn a mapping between the output neurons of the TDE units and the classification neurons representing the two-speed classes, a triplet STDP learning rule is applied. In this project, the implementation presented in [29] is used. During the learning process, an array of spike times is randomly selected from a collected data set.…”

“…Zhao et al demonstrate how a PID controller can be implemented with relational networks on the DynapSE-1 processor to control a robotic arm. [25]. One essential component of their proposed controller is a relational network which links two 1D input populations to a 1D output population via a 2D hidden layer of neurons.…”