Electronic spiking neuron
Enabling & Support

A robot car solved a maze using spike neural networks

11/12/2019 95 views 1 likes
ESA / Enabling & Support / Space Engineering & Technology / Shaping the Future

By using novel computing software that mimics the nervous system a TDE activity has developed a powerful new hardware that opens up the possibility for versatile, fast, low power and noise immune alternatives to state-of-the-art digital processors. The activity tested their processor by building a car capable of solving a simple labyrinth.

The activity, with UNINOVA - CTS and DEIMOS Engenharia in Portugal, wanted to demonstrate how using a new coding technique called spike trains, where a sequence of binary numbers are fed into the processor, a train of zeros and ones, that correspond to ‘spikes’ on or off, could be used to support missions. These spike trains are meant to mimic neuron pathways, which can create an action potential, the spike, a brief electric pulse with an approximate duration of 1 ms. In artificial neuromorphic computing artificial neurons are designed and implemented in software. Non-spiking neural networks have intrinsic limitations, mainly for processing large amount of data or for fast adaptation in an uncertain environment.

Almost all space activities require a huge amount of computer processing, from data analysis to coding the paths of rovers to launching spacecraft. This new spike method could mean that all of these activities are able to cope with larger amounts of information or conduct the processes faster.

Neuromorphic spike controlled car
Neuromorphic spike controlled car

The new technique means lower power bit coding schemes for control and data processing apparatus, more robust computing and communication in the harsh manufacturing or Space environments and can be used on or off board to support any activity before, during and after any Space or commercial mission.

The activity tested their new technique by operating a small car using the spike computing method. The goal was to develop a robot that would learn how to solve a maze. The car underwent two phases, one where it made mistakes as it learnt how to solve the maze, then another where it was expected to complete the maze by following the correct path from memory every time.

Ultimately, the activity found that in order to take advantages of the benefits of spike computing the devices built need to guarantee good voltage spectral characteristics without incurring unforeseen behaviour elicited by the short-time spike energy.

TDE contract 4000117067 closed in November 2019