A low-power, wireless, hand-mounted, pointing and arm gesture recognition tool has been designed for convenient, screen-free, symbolic communication for mobile robots in the field. The device is based on commercially-available integrated sensors (accelerometer, gyroscope, and magnetometer), an IoT-enabled microcontroller, and 3-D printed parts. The design of the gesture set dramatically simplifies recognition and operator training and reduces power consumption. Guided by cognitive load and power considerations, the device is triggered by finger switches but does not significantly impede other manual tasks. We describe and analyze the design of the device and gestures and demonstrate the recognition performance on a set of 12 gestures using 3 test subjects. Easily calibrated, trained, and reconfigured, the system can achieve essentially 100% recognition when operating on “self”-trained gestures.

Signal representation and computational principles discovered in biological neural networks have long challenged our assumptions on how to build artificial robotic systems. In the context of spatial navigation, this work assumes that `place' cells, found in the mammalian hippocampus, are involved in providing directions to a distant, but previously visited goal. This work uses a sparse topological map representation of the environment using synaptically-linked place cells, a more biologically-plausible representation than dense occupancy grids. Previous work has shown the plausibility of using spike-propagation in this network to perform path planning. In this work, we show that biologically-realistic addition of coincident spikes can speed up propagation that favors paths that provide redundancy. This preference can be interpreted as risk aversion, offering a lower average cost of travel, when sections of the planned path are discovered to be non-traversable. We demonstrate some of these properties on our model neuromorphic VLSI system.

Neurons in the medial entorhinal cortex of rats have been found to respond in a two-dimensional hexagonal “grid” pattern anchored to the environment. “Grid cells” with different spatial frequencies are thought to contribute to the creation of unimodal “place” cell responses useful for spatial navigation. In this paper we present results from an analog VLSI circuit that generates a hexagonal grid of activity using continuous attractor dynamics and transmits this pattern via neuron-like spikes. This circuit is a component of a larger system for modeling the neural circuits underlying mammalian spatial navigation.

Many animals are known to maintain an internal estimate of their orientation in the environment. In the absence of external sensory cues, this estimate inevitably exhibits drift. When sensory information is available, associations between sensory landmarks and the internal estimate can be used to correct for drift. In this paper we present a neuromorphic system to model such associations between sensory landmarks in the environment (as provided by sonar) and the activity of a hardware-based head direction cell system (HDS) that continuously integrates angular velocity signals to maintain an estimate of the orientation. These associations are shown to correct for drift errors that are encountered in the HDS.

The lateral superior olive is the first neural area to receive binaural inputs in the mammalian sound localization pathway. Selective to interaural level differences (ILD) while intensity-independent, these neurons signal when a sound arrives from within a limited range of angles. Although several different computational models have been proposed to explain ILD-selectivity, we highlight one that has the greatest utility in bat echolocation. We argue for a detection mechanism based on the ratio of excitatory and inhibitory synaptic conductances. An analog VLSI neuron performing this computation in an echolocation system is demonstrated with realistic echolocation signals.

The head direction (HD) cell system in the brain of mammals is thought to be part of the neural circuitry supporting their spatial navigation capabilities. In this paper we present a neuromorphic VLSI system that mimics the operation of the HD system. Relying on spiking neurons and attractor dynamics, this system can be used to represent and retain the current estimate of head orientation in the environment and integrate angular velocity to update this estimate. The presented system is a first step towards more general odometry systems that could be used for modeling biological navigation systems.

To support our ongoing work in modeling bat echolocation, an artificial bat head was designed and fabricated using a 3D printer, an ultrasonic cochlea-like filter bank with 16 channels was designed with moderate quality (Q) factor, and 128 spiking neurons convert these signals to spike trains. A two-dimensional address-event arbiter is used to transmit these spikes off of the chip. We demonstrate that the population of spiking neurons can be decoded to estimate azimuth and elevation of ultrasonic chirps. This chip was fabricated in a commercially-available 0.5μm CMOS process and consumes approximately 36μW.

In this paper, we describe a method for transient stochastic analysis and apply it to develop a stochastic model for a true random number generator (RNG) circuit using intrinsic circuit noise. We use numerical simulation of stochastic differential equations to obtain time-domain transient analysis of the circuit. The simulation shows similar stochastic behavior and probability tuning as that observed in measurements of the fabricated chips. We further develop a small signal stochastic model of the circuit. The model reveals the role each device plays in contributing to overall stochastic behavior. The model and the simulation allow us to predict how the device parameters affect the performance. The method we propose here can be applied to other circuits where stochastic sample paths and ensemble statistics are necessary to characterize the circuits. Traditional noise analysis in the frequency domain is not adequate to provide this information.

We report a compact realization of short-term depression (STD) in a VLSI stochastic synapse. The behavior of the circuit is based on a subtractive single release model of STD. Experimental results agree well with simulation and exhibit expected STD behavior: the transmitted spike train has negative autocorrelation and lower power spectral density at low frequencies which can remove redundancy in the input spike train, and the mean transmission probability is inversely proportional to the input spike rate which has been suggested as an automatic gain control mechanism in neural systems. The dynamic stochastic synapse could potentially be a powerful addition to existing deterministic VLSI spiking neural systems.

Echolocating bats have long demonstrated the potential for air-coupled sonar to guide agile navigation through complex 3D environments. While air-coupled sonar has been extensively used in mobile ground robotics and a considerable amount of the bat's auditory system has been studied, little work has been done to understand how the neural circuitry of the bat can integrate echoes into a coherent, behaving system. This paper addresses the unique characteristics of bat echolocation and how they impact the design of a neuromorphic VLSI-based echolocation system. By constructing a system to solve bat-specific sensorimotor tasks the authors have encountered a number of interesting issues and questions that commonly face neuromorphic VLSI designers. Successes, problematic issues, and future directions are discussed

Integrated CMOS neural amplifiers have recently grown in importance as large microelectrode arrays have begun to be practical. We previously reported low-power neural amplifiers with integrated pre-filtering and measurements of the spike signal to reduce data bandwidth and to facilitate spike-sorting prior to transmission to a data-acquisition system. Characteristics of a prototype circuit were reported using a 1.5 V power supply, suitable for single cell battery operation. Here we report improved transient amplitude tracking as well as application of the circuit in live recordings of wide-field motion sensitive cells in blowflies. The features extracted by the chip efficiently discriminate between stimulus conditions, as demonstrated by receiver operating characteristic analysis.

The rapid control of sonar-guided vehicles through obstacle fields has been a goal of robotics for decades. How sensory data is represented strongly affects how obstacles and goal information can be combined to select a direction of travel. While typical approaches combine attractive and repulsive effects to directly determine steering, we are investigating an algorithm that evaluates multiple directions simultaneously followed by a winner-take-all (WTA) function which then guides steering. In this paper we describe a neuromorphic VLSI implementation of this algorithm using the inherent echo delay to create a range-dependent gain in a 'race-to-first-spike' neural WTA circuit. The chip was fabricated in a commercially-available 0.5 mum CMOS process and in this paper we present preliminary test results

Synapses are a critical element of biologically-realistic, spike-based neu- ral computation, serving the role of communication, computation, and modification. Many different circuit implementations of synapse func- tion exist with different computational goals in mind. In this paper we describe a new CMOS synapse design that separately controls quiescent leak current, synaptic gain, and time-constant of decay. This circuit im- plements part of a commonly-used kinetic model of synaptic conduc- tance. We show a theoretical analysis and experimental data for proto- types fabricated in a commercially-available 1.5µm CMOS process.

One way to understand a neurobiological system is by building a simulacrum that replicates its behavior in real time using similar constraints. Analog very large-scale integrated (VLSI) electronic circuit technology provides such an enabling technology. We here describe a neuromorphic system that is part of a long-term effort to understand the primate oculomotor system. It requires both fast sensory processing and fast motor control to interact with the world. A one-dimensional hardware model of the primate eye has been built that simulates the physical dynamics of the biological system. It is driven by two different analog VLSI chips, one mimicking cortical visual processing for target selection and tracking and another modeling brain stem circuits that drive the eye muscles. Our oculomotor plant demonstrates both smooth pursuit movements, driven by a retinal velocity error signal, and saccadic eye movements, controlled by retinal position error, and can reproduce several behavioral, stimulation, lesion, and adaptation experiments performed on primates.

The localization and orientation to various novel or interesting events in the environment is a critical sensorimotor ability in all animals, predator or prey. In mammals, the superior colliculus (SC) plays a major role in this behavior, the deeper layers ex(cid:173) hibiting topographically mapped responses to visual, auditory, and somatosensory stimuli. Sensory information arriving from differ(cid:173) ent modalities should then be represented in the same coordinate frame. Auditory cues, in particular, are thought to be computed in head-based coordinates which must then be transformed to reti(cid:173) nal coordinates. In this paper, an analog VLSI implementation for auditory localization in the azimuthal plane is described which ex(cid:173) tends the architecture proposed for the barn owl to a primate eye movement system where further transformation is required. This transformation is intended to model the projection in primates from auditory cortical areas to the deeper layers of the primate superior colliculus. This system is interfaced with an analog VLSI-based saccadic eye movement system also being constructed in our labo(cid:173) ratory.