Projects
Uncertainty Quantification in Model Based Deep Learning
Various algorithms combine deep neural networks (DNNs) and Kalman filters (KFs) to learn from data to track in complex dynamics. Unlike classic KFs, DNN-based systems do not naturally provide the error covariance alongside their estimate, which is of great importance in some applications, e.g., navigation. To bridge this gap, in this work we study error covariance extraction in DNN-aided KFs. We examine three main approaches that are distinguished by the ability to associate internal features with meaningful KF quantities such as the Kalman gain (KG) and prior covariance. We identify the differences between these approaches in their requirements and their effect on the training of the system. Our numerical study demonstrates that the above approaches allow DNN-aided KFs to extract error covariance, with most accurate error prediction provided by model-based/data-driven designs.
Multiple Target Tracking: Revealing Causal Interactions in Complex Systems
Revealing the underlying interaction in an observed physical system is a long-time challenge tackled by many [1]. Recently, some attempts were made to mathematically define the causality in physical systems. This work suggests estimating the forecasted states of detected objects using a mathematical time series models and methods, together with a two-layer estimation algorithm: one for the targets’ location and another for the dependence of their trajectories. Some theoretical basis will be introduced followed by a simulation of artificially generated bird flocking move ment. Some results will be shown together with conclusions.
DNN Aided Step Detection
In robotics, gait planning relies heavily on the knowledge whether each foot touches the ground. For that purpose, sensors are placed on each leg tip, indicating whether or not a step occurred. Being able to detect a step by using the IMU measurements, might allow redundancy of the force sensor (or any step detection sensor) in robotics -- replacing hardware with a software algorithm.
When a pedestrian walks, its foot accelerates in a repetitive, periodic pattern. This project assumes that each part of the period indicates a different phase of the walk, i.e., lift foot, swing foot, step etc. Using recorded and labelled data from a pedestrian walk, this project aims to predict whether or not a step occurs - using IMU (Inertial Measurement Unit) measurements only.
By looking at the future contribution of this project (in another domain), being able to detect and count steps could lead to the ability of tracking a user's healthcare of any person having a smartwatch or any other IMU including device.
Stroofie
The Stroofie is a drug detection reusable device that sits on the rim of a glass or cup.
Inside this device is a piece of gabazine – a chemical type of paper, that sits on an electric conductor. The fumes of the drink are enough to set off a chemical reaction in the gabazine, meaning the Stroofie never had to touch the drink, and thus works "automatically".
In the case of date rape drugs, if the gabazine senses them in the fumes of the drink it creates a chemical reaction, which then on the conductor closes an electrical circuit and a light turns on warning you your drink has been drugged.
While the Stroofie is reusable, the gabazine is not, meaning customers will need to place a new piece of gabazine in the Stroofie device in cases where the drink has been drugged. The development of the gabazine and the placement of it on a conductor was done in collaboration with The BGU Nanotechnology Sensors Probiotics Lab.