Working in the intersection of control, computation, communication and sensing

Synchronous Rendezvous in Geophysical Flows

Fri, 6 May 2016 17:16:54 -0400

More than 70% of the Earth’s surface is covered in water. There is a great need to survey, monitor, and search over large swaths of ocean, either for monitor algae, measure water quality, or track the dissipation of pollutants. Miniature, inexpensive, semi-passive drifters can be deployed in numbers to form a large-scale sensor network over an aquatic environment for survey purposes, but in order to maximize their deployment time, given their limited resources, it makes sense to use the natural current and “ride” this flow as they make their way through the ocean. How the floaters can make the most use of this ambient resource and harvest this energy from their physical environment to perform their survey task is one big part of this collaborative research effort.

At the same time, the limited storage and communication capabilities of the drifters require novel solutions to the data management problem: can we use more capable autonomous vehicles to either heard or rendezvous with the drifters to allow the latter to upload their data “catch?” How do the drifters need to coordinate between themselves and with their “data mules” to maximize the efficiency of the data transfer and optimize the deployment of the overall network?

Mobility-enabled nuclear detection

Thu, 5 May 2016 17:40:31 -0400

Mobile radiation detection networks can offer an additional layer of safety and security against the release and/or proliferation or radiological agents. For these detector networks to scale, the sensors themselves have to be relatively inexpensive, so that they can be deployed in numbers. Using commercial off-the-shelf detector technology, this project investigates how controlled sensor mobility and judicious networking can help boosting the detection performance of such networks, especially in cases where a) the signal that has to be detected is weak and comparable to background noise, b) it is in transit and therefore the detection time window is limited, and c) the mobile detectors operate in environments where accurate localization may not be possible. In this project, we will be exploring the possibility of equipping small aerial vehicles with miniature radiation detectors and steering them into tracking, intercepting, and characterizing mobile targets within constrained, GPS-denied environments.

GEAR

Sun, 25 Oct 2015 00:00:00 -0400

GEAR is a collaborative research effort between the University of Delaware and Johns Hopkins University that brings together robotics engineers, cognitives scientists, and physical therapists, for the purpose of designing new rehabilitation environments and methods for young children with mobility disorders.

The envisioned pediatric rehabilitation environment consists of a portable harness system intended to partially compensate for body weight and facilitate the children’s mobility within a 10 x 10 feet area, a small humanoid robot that socially interact with subjects trying to engage with them in games designed to make them maintain particular levels of physical activity, and a network of cameras capturing and identifying the motion in the environment and informing the robot so that the latter adjusts its behavior depending on that of the child.

The realization of this system presents unique new research challenges in the fields of pediatric rehabilitation, robot control, machine vision, and computational learning.

sensor management for decision making

Sat, 1 Aug 2015 09:01:46 -0400

The optimal method for the detection of a signal generated by a stochastic process involves a binary hypothesis test in which a certain statistic is calculated based on observations, and then the resulting number is compared to a fixed threshold. Depending on whether the number is above or below the threshold, one hypothesis is more likely than the other. This mathematical process inherently accounts for the randomness of the observation which is due to the process that generates the observation, in other words, it implicitly assumes that the measurement of the observation itself is perfect.

In the context of remotely detecting radioactive sources using mobile sensors, one has to make a similar binary decision: is a source present or not? A binary hypothesis test of this nature can be used to decide between the two cases. However, when the observations of radioactivity are made by sensors which themselves are moving with some degree of randomness, it is important to be able to assess how the accuracy of the final decision may degrade as a result of this random sensor motion.

This material is based upon work supported by the National Science Foundation under Grant No. 1548149. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Control and learning

Thu, 15 Jul 2010 13:37:31 -0400

Collaborative work between University of Delaware and Boston University

Autonomous agents capable of interacting with their physical environment are subject to constraints in terms of what actions they can perform after certain other actions. For example, a robot that is holding a pencil in its gripper cannot pick up another object with this same gripper before dropping the one that is currently holding. These type of logical constraints, impose interdependencies between the tasks or the actions that an agent is capable of doing.

Similar type of constraints we see in natural languages, and specifically in phonology. In any natural language, there are some sounds (phonems) that naturally come after other sounds in the words of the language. Other sounds simply don’t go together -they don’t sound right. The key insight in this project is to adapt efficient models and methods that computational linguists are using to capture the early acquisition of language and transfer them over to the field of hybrid dynamical systems. We develop new theories for planning, control synthesis, and learning in engineered systems that interact with each other and with their environment. Our current approach combines elements of game theory with grammatical inference and hybrid systems abstraction to produce new symbolic control synthesis methods.

The following is a video of our implementation of our algorithms for grammatical inference and symbolic LTL control synthesis.

Code (in python) that implement our learning algorithms and carry out the special operations of (semi)automata composition in the above demo can be found at : https://github.com/prarobo/cps_multi_agent

Check out also the related NSF Science Nation article.

This material is based upon work supported by the National Science Foundation under Grant No. 1035577. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Hearing the shape of a drum

Wed, 15 Jul 2009 13:47:29 -0400

Hearing the “shape of a drum” refers to a long standing mathematical problem of determining a drum’s membrane geometry from its vibration modes. This problem relates to applications in pattern recognition and shape identification. In 1992, it was shown that the shape of the drum cannot be uniquely determined by its vibration modes, but some encouraging results suggest that certain combinations of the Dirichlet eigenvalues can provide reliable, scaling and rotation invariant descriptors for a planar curve.

This project builds on this idea and suggests new ways in which formations of robots can collect data, implement a spatially distributed neural network on their on-board processors, and autonomously classify what “they are looking at” collectively, without any one of them having to piece together the whole picture and perform the computation alone. The distributed approach to motion control, data collection, and processing gives inherent robustness both against measurement noise and physical failures, and does not necessitate human involvement in the interpretation of the data.

A more comprehensive Final Project Report

This material is based upon work supported by the National Science Foundation under Grant No. 0822845. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Micro-vehicle autonomy

Sat, 1 Mar 2008 16:20:03 -0500

We work within a consortium of academic and industry groups to develop heterogeneous networked groups of small vehicles and sensors operating in dynamic, resource-constrained, adversarial environments. These platforms are intended for urban reconnaissance and surveillance, are supposed to provide situational awareness, and are expected to be able to operate under significant uncertainty. Our own specific contribution is in modeling, abstraction, and motion control of these mobile platforms, paying special attention to the limited power, computational, and sensory resources that these vehicles are envisioned to have.

The approach we follow to navigation and control is based on a novel formulation of receding horizon optimization, specifically designed for computational platforms with limited power and without reliance to sophisticated numerical optimization libraries. In this context, we are studying an array of low-dimensional hybrid models of progressing complexity, including fully deterministic, uncertain, and finally stochastic continuous dynamics.

Check our movies how we enhance miniature mobile robot capabilities.