This work was developed by F. Slijkhuis, supervised by S. Keemink and P. Lanillos as part of the SPIKEFERENCE project, co-founded by the Human Brain Project (HBP) Specific Grant Agreement 3 (ID: 945539) and the Donders Institute for Brain, Cognition and Behaviour.
Slijkhuis, F. S., Keemink, S. W., & Lanillos, P. (2023). Closed-form control with spike coding networks. IEEE Tran. Cog. Dev. Sys. SI Advancing AI with neuromorphic computing.
We develop the Spiking Coding Network theory for estimation and control. The resulting networks work as a spiking equivalent of a linear–quadratic–Gaussian controller. We demonstrate robust spiking control of simulated spring-mass-damper and cart-pole systems, in the face of several perturbations, including input- and system-noise, system disturbances, and neural silencing. As our approach does not need learning or optimization, it offers opportunities for deploying fast and efficient task-specific on-chip spiking controllers with biologically realistic activity.
(A) SMD, (B) Estimation of the SMD and (C) Control of the SMD and study of robustness against neural silencing.
(A) Cartpole model, (B) Estimation and Control with a step function as reference (Spiking LQG)
