A new mem-computing strategy for fast low-power robot motion control


Prof. Alon Ascoli, Technology University of Dresden, Germany-- 26-02-2018


The massively-parallel computing power of biological systems, endowed with cells capable to store and process data in the same infinitesimal volumes, lies at the origin for their time and energy efficiency. The computing paradigms implemented in future electronic systems shall be inspired to the operating principles of biological systems, allowing to enhance considerably the speed and power consumption of traditional machines.

Memristors are the circuit elements which best mimic the nonlinear dynamics of biological systems, given that some real-world nano-devices definable as such may undergo smooth resistance changes upon variations in the history of the stimulus applied to them, or keep their current resistance indefinitely under zero input, similarly as neural synapses, while others, capable to amplify infinitesimal fluctuations of energy, may induce the generation of a threshold driven spike with signal gain, resembling an action potential emerging from the axon of a neuron, in a simple circuit, which would otherwise have limited data processing functionalities. On the basis of these considerations, the adoption of memristors in circuit and system design is expected to enhance the performance of state-of-the-art data processors.

In a recent study [1] we demonstrated that, replacing linear resistors with hafnium oxide  memristors within the cells of standard Cellular Nonlinear Networks (CNNs), the resulting nonlinear arrays may process data leveraging the dynamics of the cells’ capacitor voltages, as in the original computing structures, but may also store computation results as memristance values. This may allow to address the main issue which designers are presently facing in the development of analog electronics-based implementations of intelligent visual sensors processing information according to the CNN computing paradigm, i.e. the significant mismatch between the high resolution of state-of-the-art CMOS image sensor arrays and the limited maximum number of signal processing elements which may possibly fit within the available integrated circuit area in the latest hardware realizations of cellular networks.

In this presentation we provide yet another example for the great potential of memristors in electronics. The current electro-mechanical control system enabling the motion of a limb of a humanoid robot called Myon from the stable rest state to the unstable upright position is based upon a power-hungry and time-consuming iterative cycle of sense and drive phases, known as Go-Against-the-Force (GAF) protocol, in which a motor respectively measures the angular velocity of the falling limb subject to gravitation, allowing its simultaneous inverting time integration, and applies a torque proportional to the value of this integral at the end of the previous phase to the limb under control so as to counteract and prevail over the gravitation torque inducing its expected lift.

In order to increase the speed of the original control action while reducing its energy costs, we propose a new protocol [2], called Kick-Fly-Catch (KFC), inspired to the Maximum Principle of Pontryagin, and consisting of three main phases: the first one, where the largest possible torque is applied to the limb inducing a significant increase in its height, the second one, where the limb is left free to continue its ascent leveraging the momentum gained in the Kick phase, and the third one, initiated when the limb angular velocity changes polarity or increases in modulus, where the application of the original GAF protocol allows to position and maintain the limb around the unstable upright position. Provided that the time duration of the Kick phase is properly calibrated so that the limb quickly attains a position close to the target destination before the Fly phase is  commenced, given that the slow and energy-wasting Catch Phase would then be deployed only to stabilize the upright state, the proposed control strategy outperforms the original approach in terms of time and energy efficiency. The unique mem-computing capabilities of a passive nonvolatile memristor may be harnessed to determine an optimal estimate for the time duration of the Kick phase, store its value when the control system is in sleep mode, and correct it upon changes to the nominal operating conditions. LTSpice simulations on circuit-theoretic implementations of the overall electromechanical system under the GAF and KFC control protocols [3] validate the theoretical predictions, showing the significant benefits of the proposed strategy over the standard one in relation to time and energy efficiency without degradation on the level of adaptability to perturbations to the nominal operating conditions.


[1] A. Ascoli, R. Tetzlaff, D. Ielmini, and L.O. Chua, “Cellular Nonlinear Networks with real-world memristors: a paradigm for memcomputing”, 5th Workshop on Memristor Technology, Design, Automation, and Computing (mDAC), HiPEAC, Manchester, 24th Jan. 2018

[2] A. Ascoli, D. Baumann, R. Tetzlaff, L.O. Chua, and M. Hild, “Memristor-enhanced humanoid robot control system-Part I: theory behind the novel memcomputing paradigm”, Int. Journal of Circuit Theory and Applications (IJCTA), 2017, DOI: 10.1002/cta.2431

[3] D. Baumann, A. Ascoli, R. Tetzlaff, L.O. Chua, and M. Hild, “Memristor-enhanced humanoid robot control system-Part II: circuit theoretic model and performance analysis”, Int. Journal of Circuit Theory and Applications (IJCTA), 2017, DOI: 10.1002/cta.2430


Alon Ascoli received a Ph.D. Degree in Electronic Engineering from University College Dublin in 2006. From 2006 to 2009 he worked as RFIC analog engineer at CSR Sweden AB. From 2009 to 2012 he was Research Assistant in the Department of Electronics and Telecommunications at Politecnico di Torino. Since 2012 he is Assistant Professor in the Faculty of Electrical and Computer Engineering, Technische Universität Dresden. His research interests lie in the area of nonlinear circuits and systems, networks of oscillators, Cellular Nonlinear Networks and memristors. Dr. Ascoli was honored with the IJCTA 2007 Best Paper Award. In April 2017 he was conferred the habilitation title as Associate Professor in Electrical Circuit Theory from the Italian Ministry of Education. In November 2017 he was conferred a High-Performance Award from Technische Universität Dresden. Since 2014 he is Management Committee Substitute for Germany in the COST Action IC1401 MemoCIS “Memristors – Devices, Models, Circuits, Systems, and Applications”. He has been Program Chair and Special Session Chair for the 15th International Workshop on Cellular Nanoscale Networks and their Applications (CNNA) in 2016.




CE Tweets