Chiara Bartolozzi

Asynchronous frameless event-based optical flow

This paper introduces a process to compute optical flow using an asynchronous event-based retina at high speed and low computational load. A new generation of artificial vision sensors has now started to rely on biologically inspired designs for light acquisition. Biological retinas, and their artificial counterparts, are totally asynchronous and data driven and rely on a paradigm of light acquisition radically different from most of the currently used frame-grabber technologies. This paper introduces a framework for processing visual data using asynchronous event-based acquisition, providing a method for the evaluation of optical flow. The paper shows that current limitations of optical flow computation can be overcome by using event-based visual acquisition, where high data sparseness and high temporal resolution permit the computation of optical flow with micro-second accuracy and at very low computational cost.

Global scaling of synaptic efficacy: Homeostasis in silicon synapses

Synaptic homeostasis is a mechanism present in biological neural systems that acts to maintain an homogeneous and stable computational substrate, in face of intrinsic inhomogeneities among neurons, and of their continuous changes due to learning processes and variations in the statistics of the input signals. In hardware spike-based neural networks homeostasis could be useful for solving issues such as mismatch and temperature drifts. Here we present a synaptic circuit that supports both spike-based learning and homeostatic mechanisms, and show how it can be used in conjunction with a software control algorithm to model global synaptic scaling homeostatic mechanism.

Embedded neuromorphic vision for humanoid robots

We are developing an embedded vision system for the humanoid robot iCub, inspired by the biology of the mammalian visual system, including concepts such as stimulus-driven, asynchronous signal sensing and processing. It comprises stimulus-driven sensors, a dedicated embedded processor and an event-based software infrastructure for processing visual stimuli. These components are integrated with the existing standard machine vision modules currently implemented on the robot, in a configuration that exploits the best features of both: the high resolution, color, frame-based vision and the neuromorphic low redundancy, wide dynamic range and high temporal resolution event-based sensors. This approach seeks to combine various styles of vision hardware with sensorimotor systems to complement and extend the current state-of-the art.

Asynchronous visual event-based time-to-contact

Reliable and fast sensing of the environment is a fundamental requirement for autonomous mobile robotic platforms. Unfortunately, the frame-based acquisition paradigm at the basis of main stream artificial perceptive systems is limited by low temporal dynamics and redundant data flow, leading to high computational costs. Hence, conventional sensing and relative computation are obviously incompatible with the design of high speed sensor-based reactive control for mobile applications, that pose strict limits on energy consumption and computational load. This paper introduces a fast obstacle avoidance method based on the output of an asynchronous event-based time encoded imaging sensor. The proposed method relies on an event-based Time To Contact (TTC) computation based on visual event-based motion flows. The approach is event-based in the sense that every incoming event adds to the computation process thus allowing fast avoidance responses. The method is validated indoor on a mobile robot, comparing the event-based TTC with a laser range finder TTC, showing that event-based sensing offers new perspectives for mobile robotics sensing.

Event-driven visual attention for the humanoid robot iCub.

Fast reaction to sudden and potentially interesting stimuli is a crucial feature for safe and reliable interaction with the environment. Here we present a biologically inspired attention system developed for the humanoid robot iCub. It is based on input from unconventional event-driven vision sensors and an efficient computational method. The resulting system shows low-latency and fast determination of the location of the focus of attention. The performance is benchmarked against an instance of the state of the art in robotics artificial attention system used in robotics. Results show that the proposed system is two orders of magnitude faster that the benchmark in selecting a new stimulus to attend.

An Asynchronous Neuromorphic Event-Driven Visual Part-Based Shape Tracking

Object tracking is an important step in many artificial vision tasks. The current state-of-the-art implementations remain too computationally demanding for the problem to be solved in real time with high dynamics. This paper presents a novel real-time method for visual part-based tracking of complex objects from the output of an asynchronous event-based camera. This paper extends the pictorial structures model introduced by Fischler and Elschlager 40 years ago and introduces a new formulation of the problem, allowing the dynamic processing of visual input in real time at high temporal resolution using a conventional PC. It relies on the concept of representing an object as a set of basic elements linked by springs. These basic elements consist of simple trackers capable of successfully tracking a target with an ellipse-like shape at several kilohertz on a conventional computer. For each incoming event, the method updates the elastic connections established between the trackers and guarantees a desired geometric structure corresponding to the tracked object in real time. This introduces a high temporal elasticity to adapt to projective deformations of the tracked object in the focal plane. The elastic energy of this virtual mechanical system provides a quality criterion for tracking and can be used to determine whether the measured deformations are caused by the perspective projection of the perceived object or by occlusions. Experiments on real-world data show the robustness of the method in the context of dynamic face tracking.

Asynchronous, event-driven readout of POSFET devices for tactile sensing

In this work, we report a novel circuit architecture to implement event-driven tactile sensing using the POSFET tactile device. The proposed circuit matches advantages of the POSFET device (integration of sensing and electronics on the same die, high electromechanical transduction bandwidth, etc.) with the ones of the event-driven approach. In the proposed circuit, the POSFET device is interfaced with a spiking neuron, of the type integrate and fire: the input mechanical stimulus is translated into digital pulses. The proposed approach paves the way for the implementation of neuromorphic integrated tactile sensing systems based on POSFET devices.

Implementing homeostatic plasticity in VLSI networks of spiking neurons

Homeostatic plasticity acts to stabilize firing activity in neural systems, ensuring a homogeneous computational substrate despite the inherent differences among neurons and their continuous change. These types of mechanisms are extremely relevant for any physical implementation of neural systems. They can be used in VLSI pulse-based neural networks to automatically adapt to chronic input changes, device mismatch, as well as slow systematic changes in the circuitpsilas functionality (e.g. due to temperature drifts). In this paper we propose analog circuits for implementing homeostatic plasticity mechanisms in VLSI spiking neural networks, compatible with local spike-based learning mechanisms. We show experimental results where a homeostatic control is implemented as a hybrid SoftWare/HardWare (SW/HW) solution, and present analog circuits for a complete on-chip stand-alone solution, validated by circuit simulations.

Ultra Low Leakage Synaptic Scaling Circuits for Implementing Homeostatic Plasticity in Neuromorphic Architectures

Homeostatic plasticity is a property of biological neural circuits that stabilizes their neuronal firing rates in face of input changes or environmental variations. Synaptic scaling is a particular homeostatic mechanism that acts at the level of the single neuron over long time scales, by changing the gain of all its afferent synapses to maintain the neurons mean firing within proper operating bounds. In this paper we present ultra low leakage analog circuits that allow the integration of compact integrated filters in multi-neuron chips, able to achieve time constants of the order of hundreds of seconds, and describe automatic gain control circuits that when interfaced to neuromorphic neuron and synapse circuits implement faithful models of biologically realistic synaptic scaling mechanisms. We present simulation results of the low leakage circuits and describe the control circuits that have been designed for a neuromorphic multi-neuron chip, fabricated using a standard 180nm CMOS process.

Event-driven embodied system for feature extraction and object recognition in robotic applications

A major challenge in robotic applications is the interaction with a dynamic environment and humans which is typically constrained by the capability of visual sensors and the computational cost of signal processing algorithms. Addressing this problem the paper presents an event-driven based embodied system for feature extraction and object recognition as a novel efficient sensory approach in robotic applications. The system is established for a mobile humanoid robot which provides the infrastructure for interfacing asynchronous vision sensors with the processing unit of the robot. By applying event-feature ”mapping” the address event representation of the sensors is enhanced by additional information that can be used for object recognition. The system is presented in the context of an exemplary application in which the robot has to detect and grasp a ball in an arbitrary state of motion.