Benchmark and Applications
AutoVAPS is an IoT-Enabled public safety service which integrates body-worn cameras and other sensors on the vehicle for public safety. In AutoVAPS, we propose a reference architecture that consists of the data layer for data management, the model layer for edge intelligence, and the access layer for privacy-preserving data sharing and access. Object detection is implemented as a case study of AutoVAPS. (NIST Imagery Dataset for Public Safety)
CAVBench is a benchmark suite to evaluate the connected and autonomous vehicles (CAVs) computing platforms performance. It takes six diverse realworld on-vehicle applications as evaluation workloads. We collect four real-world datasets for CAVBench as the standard input to the six application including three types data: image, audio, and text. And CAVBench has two categories of output metrics, the execution time breakdown and service-resource utilization curve (QoS-RU curve) for each application.
In this project, we proposed a synchronized data illustration and collection method to assist the data processing applications for autonomous driving. Our proposed solution can be directly deployed on autonomous vehicles for data integration and environment analysis to support the driving model construction. The experimental results validate that our proposed method can present a 360° synchronized view while providing the capability of real-time scanning with up to 80% reduced latency.
pBEAM is a collaborative cloud-edge computation system for personalized driving behavior modeling. The driving behavior model is built on top of Generative Adversarial Recurrent Neural Networks, which adapts to the dynamic change of normal driving. Transfer learning from cloud to edge improves the model performance and robustness on the edge. Experimental results on driving data from both real world and driving simulator show that the proposed algorithm achieves the best performance among all the methods.
An edge-based attack detection in ridesharing services, namely SafeShareRide, which can detect dangerous events happening in the vehicle in near real time. The detection of SafeShareRide consists of three stages: speech recognition, driving behavior detection, and video capture and analysis. We implemented SafeShareRide system by leveraging open source algorithms.
This research investigates how to deploy advanced machine learning algorithms, especially deep learning algorithms on heterogeneous resource-constrained edge devices, such as the cluster of Intel Fog Reference Design and NVIDIA Jetson TX2. It further studies efficient communication protocols on multiple edge devices and strategies to support accuracy, performance, personalization, and privacy. The developed framework will support collaborative and federated learning on a cluster with dynamical joining and leaving of edge devices.
E2M is an energy-efficient middleware software stack for autonomous mobile robots. First, E2M regulates the access of different processes to sensor data, e.g., camera frames. Second, based on a predefined per-process performance metric (e.g., safety, accuracy) and desired target. Third, E2M coordinates the execution of the concurrent processes to maximize the total contiguous sleep time of the computing hardware for maximized energy savings.
Firework is a framework for big data processing and sharing among multiple stakeholders in collaborative edge environment, in which data are owned by multiple stakeholders, and is rarely shared due to various reasons, such as security concern and privacy issue. Thought the Firework interfaces, the programmer can easily and fastly develop the application on the edge by only implementing the functions and setting up the configuring of the application.
HydraOne is a hardware-softwareco-design platform built from scratch based on our experiencewith the requirements of edge computing research problems.We present the design and implementation details and discussthree key characteristics of HydraOne: design modulariza-tion, resource extensibility and openness, as well as functionisolation.
OpenEI, an Open Framework for Edge Intelligence, is a lightweight software platform to equip the edge with intelligent processing and data sharing capability. The goal of OpenEI is that any hardware, ranging from Raspberry Pi to a powerful Cluster, will become an intelligent edge after deploying OpenEI. Meanwhile, the EI attributes, accuracy, latency, energy, and memory footprint, will have an order of magnitude improvement comparing to the current AI algorithms running on the deep learning package.
We are developing on an open vehicular data analytics platform (OpenVDAP), which consists of four components: heterogeneous vehicle computing unit (VCU), operation system (EdgeOSv), driving data collector and integrator (DDI), and application library (libvdap).
We make a comprehensive performance comparison and analysis of several state-of-the-art deep learning packages on the edges, including TensorFlow, Caffe2, MXNet, PyTorch, and TensorFlow Lite. We focus on evaluating the latency, memory footprint and energy of these tools with four popular deep learning models on different edge devices, including MacBook, Intel FogNode, NVIDIA Jetson TX2, Raspberry Pi, and Nexus 6P.