CLOSED-LOOP NEURAL ENCODING AND ESTIMATION WITH SPIKING-NEURON POPULATIONS

Nervous systems achieve robust estimation and control in dynamic and uncertain environments by encoding sensory information through sparse, asynchronous spike trains distributed across neural populations. Rather than relying on continuous-valued sensory measurements as in most robotic systems, biological sensing and control are mediated by the coordinated spiking activity of neural populations, motivating biologically grounded estimation frameworks that operate directly on neural representations. Inspired by this principle, we develop a closed-loop state estimation framework that reconstructs task-related variables directly from spiking-neuron populations. The proposed architecture decomposes relative position and velocity signals during tracking into complementary subpopulations of Leaky Integrate-and-Fire (LIF) neurons, whose spike timings are converted into causal firing-rate estimates. These neural responses are decoded using a maximum-likelihood population estimator and subsequently fused through a Kalman filter to yield smooth estimates of the underlying tracking error suitable for feedback control. We evaluate the framework in a reference-tracking task modeled after the refuge-tracking behavior of weakly electric fish. Simulation results demonstrate that spiking-neuron populations provide sufficient information to estimate both position and velocity and enable stable closed-loop performance using a conventional proportional–derivative controller. By grounding state estimation and feedback control in biologically inspired neural encoding and decoding mechanisms, this work provides a control-theoretic perspective on how spike-based sensory representations can be transformed into actionable internal state estimates, with implications for understanding neural computation and for the design of neuromorphic sensing, active perception, and brain–machine interface systems.