In this work, we present a spiking neural network (SNN) based PID controller on a neuromorphic chip. On-chip SNNs are currently being explored in low-power AI applications. Due to potentially ultra-low power consumption, low latency, and high processing speed, on-chip SNNs are a promising tool for control of power-constrained platforms, such as Unmanned Aerial Vehicles (UAV). To obtain highly efficient and fast end-toend neuromorphic controllers, the SNN-based AI architectures must be seamlessly integrated with motor control. Towards this goal, we present here the first implementation of a fully neuromorphic PID controller. We interfaced Intel's neuromorphic research chip Loihi to a UAV, constrained to a single degree of freedom. We developed an SNN control architecture using populations of spiking neurons, in which each spike carries information about the measured, control, or error value, defined by the identity of the spiking neuron. Using this sparse code, we realize a precise PID controller. The P, I, and D gains of the controller are implemented as synaptic weights that can adapt according to an on-chip plasticity rule. In future work, these plastic synapses can be used to tune and adapt the controller autonomously.

Researchers working with neural networks have historically focused on either non-spiking neurons tractable for running on computers or more biologically plausible spiking neurons typically requiring special hardware. However, in nature homogeneous networks of neurons do not exist. Instead, spiking and non-spiking neurons cooperate, each bringing a different set of advantages. A well-researched biological example of such a mixed network is a sensorimotor pathway, responsible for mapping sensory inputs to behavioral changes. This type of pathway is also well-researched in robotics where it is applied to achieve closed-loop operation of legged robots by adapting amplitude, frequency, and phase of the motor output. In this paper we investigate how spiking and non-spiking neurons can be combined to create a sensorimotor neuron pathway capable of shaping network output based on analog input. We propose sub-threshold operation of an existing spiking neuron model to create a non-spiking neuron able to interpret analog information and communicate with spiking neurons. The validity of this methodology is confirmed through a simulation of a closed-loop amplitude regulating network inspired by the internal feedback loops found in insects for posturing. Additionally, we show that non-spiking neurons can effectively manipulate post-synaptic spiking neurons in an event-based architecture. The ability to work with mixed networks provides an opportunity for researchers to investigate new network architectures for adaptive controllers, potentially improving locomotion strategies of legged robots.

Spiking Neuronal Networks (SNNs) realized inneuromorphic hardware lead to low-power and low-latencyneuronal computing architectures. Neuromorphic computingsystems are most efficient when all of perception, decision making , and motor control are seamlessly integrated into a singleneuronal architecture that can be realized on the neuromorphichardware. Many neuronal network architectures address theperception tasks, while work on neuronal motor controllersis scarce. Here, we present an improved implementation of aneuromorphic PID controller. The controller was realized onIntel's neuromorphic research chip Loihi and its performancetested on a drone, constrained to rotate on a single axis. TheSNN controller is built using neuronal populations, in whicha single spike carries information about sensed and controlsignals. Neuronal arrays perform computation on such sparserepresentations to calculate the proportional, derivative, andintegral terms. The SNN PID controller is compared to a PIDcontroller, implemented in software, and achieves a comparableperformance, paving the way to a fully neuromorphic systemin which perception, planning, and control are realized in anon-chip SNN.

Neuromorphic computing currently relies heavily on complicated hardware design to implement asynchronous, parallel and very large-scale brain simulations. This dependency slows down the migration of biological insights into technology. It typically takes several years from idea to finished hardware and once developed the hardware is not broadly available to the community. In this contribution, we present the CloudBrain research platform, an alternative based on modern cloud computing and event stream processing technology. Typical neuromorphic design goals, such as small form factor and low power consumption, are traded for 1) no constraints on the model elements, 2) access to all events and parameters during and after the simulation, 3) online reconfiguration of the network, and 4) real-time simulation. We explain principles for how neuron, synapse and network models can be implemented and we demonstrate that our implementation can be used to control a physical robot in real-time. CloudBrain is open source and can run on commodity hardware or in the cloud, thus providing the community a new platform with a different set of features supporting research into, for example, neuron models, structural plasticity and three-factor learning.