Data-driven modeling usually suffers from data sparsity, especially for large-scale modeling for urban phenomena based on single-source urban infrastructure data under fine-grained spatial-temporal contexts.
Carpooling has long held the promise of reducing gas consumption by decreasing mileage to deliver co-riders. Although ad hoc carpools already exist in the real world through private arrangements, little research on the topic has been done. In this paper, we present the first systematic work to design, implement, and evaluate a carpool service, called coRide, in a large-scale taxicab network intended to reduce total mileage for less gas consumption.
The first video demonstrats detection of a wh-question non-manual marker. Tracked face and head is shown on left, while the right image shows the extracted spatial pyramid features. Red bars indicate detection of the wh non-manual marker, while blue bars indicate that the system detects no wh non-manual marker.
This paper introduces a new mobile sensor scheduling problem involving a single robot tasked to monitor several
The overreaching goal of CometCloud is to enable highly heterogeneous, dynamically federated computing and data platforms that can support end-to-end application workflows with diverse and dynamic changing application requirements. This is achieved through (a) autonomic on-demand federation of geographically distributed compute and data resources as needed by the application workflow, and (b) exposing the resulting software-defined federated cyberinfrastructure using elastic cloud abstractions and science-as-a-service platforms. As a result, CometCloud is able to create a nimble and dynamically programmable environment that autonomously evolves over time, adapting to changes in both the federated infrastructure and the application requirements.
DataSpaces is a programming system targeted at current large-scale systems and designed to support dynamic interaction and coordination patterns between scientific applications. DataSpaces essentially provides a semantically specialized shared-space abstraction using a set of staging nodes. This abstraction derives from the tuple-space model and can be associatively accessed by the interacting applications of a simulation workflow. DataSpaces also provides services including distributed in-memory associative object store, scalable messaging, as well as runtime mapping and scheduling of online data analysis operations.
The GreenHPC initiative at Rutgers is a research and educational initiative aiming at addressing several efforts in the intersection of energy efficiency, scalable computing and high performance computing. Key focus areas include (1) Energy efficiency of scientific data analysis pipelines at scale, (2) In-situ data analytics and co-processing at extreme scales and (3) Application-aware cross-layer power management for High Performance Computing systems .GreenHPC also acts as a forum for researchers and the educational community to exchange ideas and experiences on energy efficiency by disseminating research results, educational activities at different levels (PhD, MS, undergraduate - REU, K12 - GSET) and organizing events and editorial activities of related topics
Parasol is a green micro-datacenter partially powered by solar energy and partially cooled by "free-cooling". It comprises a small container, a set of solar panels, and batteries. The container lies on a steel structure placed on the roof of our building. The solar panels are mounted on top of the steel structure and shade the container from the sun. The container hosts two racks of energy-efficient servers (up to 160 of them) and networking equipment. The container uses free cooling whenever possible, and direct-exchange air conditioning otherwise.
Current advances in computer science and other disciplines rely on the massive computation horsepower of data parallel architectures, such as GPUs. Programming data parallel architecture is not easy, as it requires the efficient handling of data movements across the memory hierarchy of thousands of processing cores.
The goal of this activity is to understand the structure of complexity classes. Complexity classes provide the best tool currently available for understanding the computational complexity of real-world computational problems. Some of these problems are notoriously difficult, but recent progress justifies some optimism that additional useful insight about these complexity classes can be obtained.
Many state of the art algorithms for motion planning are concerned with asymptotic optimality (variants of PRM* and RRT*). One of the requirements to use these algorithms is the availability of a steering function for the system being planned for.
Probabilistic roadmap planners utilize an offline phase to build up information about the configuration space (C-space) and solve many practical motion planning problems. Traditionally, many of these planners focus on feasibility and may return paths of low quality; considerably different from the optimal ones, where path quality can be measured in terms of length, clearance, or smoothness. Smoothing can be used to improve some of these measures and algorithms exist that produce roadmaps with paths that are deformable to optimal ones. Hybridization graphs combine multiple solutions into a higher quality one that uses the best portions of each input path. These techniques, however, can be expensive for the online resolution of a query, especially when multiple queries must be answered.
This project focuses on Multi-Robot Path Planning, with the goal of providing a complete and tractable solution to such instances. The current work considers an abstraction of the traditional multi-robot path planning problem that is abstracted as computing non-colliding paths for multiple agents between their start and goal locations on a graph. Such a formulation is known to be NP-complete, indicating that a naive search to solve the problem will have exponential complexity in the number of robots.
The increasing availability of low-cost, compliant and human-friendly manipulators allows robots, such as Rethink Robotics' Baxter, to be placed in close proximity to human workers. Unlike traditional automation systems, which needed to be kept in cages, these compliant robots can share a common workspace with human workers. A clear benefit of this close proximity is the opportunity for cooperation between a human worker and an assistive robot.
The introduction of impedance control in 1985 by Neville Hogan paved the way for a safe, gentle and effective interaction between humans and machines. This interaction is ideal for rehabilitation and is epitomized in the design of manipulanda that pioneered clinical and neurological applications, the most prominent being the MIT-MANUS, developed by Hermano Igo Krebs. Following this line of research, we have recently introduced the MIT's pediatric Anklebot, an adaptive robotic device that provides an “assist-as-needed” therapy and targets ankle movements in children with neurological disorders (Michmizos et al. 2015). Since 2012, the rehabilitation robot has been used successfully in pilot studies in Pediatric Hospitals in the USA and Europe.
One of the biggest challenges in neurodevelopmental disorders today is that they remain behaviorally diagnosed disorders. Despite the large amounts of data from the autistic brain and an abundance of experimental evidences for its pathophysiology, no measures robustly depict the observed differences between an autistic and a typically developing brain. Recently, we measured functional connectivity in the somatosensory cortex in typically developing and autistic
Neurosurgeons have used electrical stimulation since the 60's to locate and distinguish specific brain areas. They soon discovered that stimulation of certain brain nuclei suppresses the symptoms of some neurological disorders. Recent efforts on patient-specific therapeutic approaches revealed the importance of computational methods in guiding deep brain stimulation (DBS), a neuromodulation treatment initially