Search

The manipulation of large objects and safe operation in the vicinity of humans are key capabilities of a general purpose domestic robotic assistant. We present the design of a soft, tactile-sensing humanoid upper-body robot and demonstrate whole-body rich-contact manipulation strategies for handling large objects. We demonstrate our hardware design philosophy for outfitting off-the-shelf hard robot arms and other components with soft tactile-sensing modules, including: (i) low-cost, cut-resistant, contact pressure localizing coverings for the arms, (ii) paws based on TRI's Soft-bubble sensors for the end effectors, and (iii) compliant force/geometry sensors for the coarse geometry sensing chest. We leverage the mechanical intelligence and tactile sensing of these modules to develop and demonstrate motion primitives for whole-body grasping. We evaluate the hardware's effectiveness in achieving grasps of varying strengths over a variety of large domestic objects. Our results demonstrate the importance of exploiting softness and tactile sensing in contact-rich manipulation strategies, as well as a path forward for whole-body force-controlled interactions with the world. (The supplemental video is available publicly at https://youtu.be/G8ZYgPRV5LY)., Comment: Research done at Toyota Research Institute. Accepted to the 5th IEEE International Conference on Soft Robotics (RoboSoft 2022). The supplemental video is available publicly at https://youtu.be/G8ZYgPRV5LY

Cloud Radio Access Network consists in deporting and pooling radio processing on shared servers for a better use of computing resources, more agile operations, and thus optimizing costs and energy performance. Some of these processes at the radio physical layer are highly demanding in terms of computing power, which implies high energy consumption. Our approach enables the automatic activation of hardware programmable acceleration resources that are better suited to these processes in terms of performance and energy efficiency. Low-Density Parity-Check Forward Error Correction functions standardized in 5th Generation of Mobile Networks are offloaded on these hardware acceleration resources. Kubernetes-based orchestration is made compatible with the use of a edge computing accelerated heterogeneous infrastructure (CPU, GPU, FPGA) that can be distributed at the edge of the operator’s network to deploy the RAN Distributed Unit. We assess the proof-of-concept implementation of our approach and evaluate cost in data offloading and energy consumption. This prefigures a full cloud orchestration for 5G access, such as recommended in the O-RAN consortium.

We introduce the first hippocampus segmentation method for a novel high resolution $(1 \times 1 \times 1$ mm3) diffusion tensor imaging (DTI) protocol acquired in 5.5 minutes at 3T. A new augmentation technique uses subsets of the DTI dataset to create mean diffusion weighted images (DWI) with plausible noise and contrast variations. The augmented DWI along with fractional anisotropy (FA) and mean diffusivity (MD) maps are used as inputs to a powerful convolutional neural network architecture. The method is evaluated for robustness using a second diffusion protocol.

Received signal strength information based dynamic frame-slotted Aloha (RSSI-DFSA) RFID is proposed to replace random slot access where idle and collided slots lead to low system efficiency. Tag coordinates are calculated from received signal strength information, and the slot is chosen based on the coordinates. Simulations show that RSSI-DFSA system efficiencies are 14.9%, 2.4%, and 2.1% higher, and the time efficiencies are 8.3%, 2.5%, and 1.8% higher than those of existing algorithms, when the path loss exponent is 4 and the standard deviations of the shadowing are 1, 2, and 3 dB.

Seizure detection is a major goal for simplifying the workflow of clinicians working on EEG records. Current algorithms can only detect seizures effectively for patients already presented to the classifier. These algorithms are hard to generalize outside the initial training set without proper regularization and fail to capture seizures from the larger population. We proposed a data processing pipeline for seizure detection on an intra-patient dataset from the world’s largest public EEG seizure corpus. We created spatially and session invariant features by forcing our networks to rely less on exact combinations of channels and signal amplitudes, but instead to learn dependencies towards seizure detection. For comparison, the baseline results without any additional regularization on a deep learning model achieved an F1 score of 0.544. By using random rearrangements of channels on each minibatch to force the network to generalize to other combinations of channels, we increased the F1 score to 0.629. By using random rescale of the data within a small range, we further increased the F1 score to 0.651 for our best model. Additionally, we applied adversarial multi-task learning and achieved similar results. We observed that session and patient specific dependencies were causing overfitting of deep neural networks, and the most overfitting models learnt features specific only to the EEG data presented. Thus, we created networks with regularization that the deep learning did not learn patient and session-specific features. We are the first to use random rearrangement, random rescale, and adversarial multitask learning to regularize intra-patient seizure detection and have increased sensitivity to 0.86 comparing to baseline study.

Idle slots cause identification inefficiency in Aloha-based RFID algorithms. An idle slot prediction and elimination technique is proposed. The algorithm is dubbed idle predicting dynamic frame-slotted Aloha (IP-DFSA). In IP-DFSA, the reader reads a slot for RN16 as well as the idle prediction bits. Using only a few prediction bits, IP-DFSA can predict and eliminate a significant number of successive idle slots following the current time slot. Whereas previous schemes only achieve 36% efficiency on average, simulation results show that IP-DFSA with 1, 2, 3, and 4 idle slot prediction bits achieves 45%, 52.5%, 56%, and 60% system efficiencies, with 83%, 86%, 88%, and 89% time efficiencies, respectively. The system efficiencies of IP-DFSA range from 9% higher to 24% higher than previous schemes for 1 prediction bit to 4 prediction bits. The number k of idle prediction bits is optimized, and it is revealed that k optimal =4 for tag numbers n$\geq$6.

Asymptotic orthogonality of channel vectors, also called favorable propagation (FP), plays a key role in massive multiple-input multiple-output (MIMO) systems, allowing linear processing to achieve optimality and maximize the information rate. Much research on the FP condition emerges under different assumptions, most of which has been proved to satisfy the FP condition for some massive MIMO implementations. A generic channel model without mutual coupling is proposed to study more general implementations than previously investigated. Two theorems establish: 1) conditions on the steering matrix alone that are sufficient to guarantee the FP condition holds and 2) that a simple condition on the steering matrix alone is sufficient to guarantee FP for all practical massive MIMO implementations. Then the uniform linear array, uniform planar array (UPA), and uniform circular array (UCA) antenna configurations are studied. Theoretical analyses indicate how close and how fast the inner product of different steering vectors converges to zero for a finite number of base station (BS) antennas, and prove that the FP condition is satisfied for the antenna configuration considered for an unlimited number of BS antennas only if the antenna spacing is not proportional to the reciprocal of the number of antennas. The analysis shows that the UCA has the shortest distance from the FP condition among the three antenna structures, while the UPA is especially suitable for massive MIMO systems for 5G owing to its good FP performance and compact physical size. Simulation results validate the theory.

CubeSats continue to proliferate and are an excellent low-cost method of remote sensing. A key piece of intelligent systems is sensory input, data storage, and data communications. With the continued miniaturization of technology, CubeSats will increase their sensory inputs with future miniaturization and enhance their robustness for autonomous operations if data and communications are secure. These futures inspire an intelligent system solution to on-orbit communications. This paper explores a dual-microprocessor approach to improve hardware cybersecurity of intelligent systems, with a view toward intensional intelligence as a means of adjudicating access to sensitive data onboard the CubeSat. With enhanced cybersecurity, Artificial Intelligent Systems (AIS) will add vital utility to otherwise vulnerable, autonomous systems. Using Systems Models-Based Thinking, we shed light on our plan to apply artificial intelligent system concepts to advance CubeSat technology. Managing technology for AIS reduces some of the uncertainties and risks associated with the space environment.

Because of speckle noise, estimating extended target scattering properties from Single Look Complex PolSAR data is a difficult problem. Speckle filtering is aiming at reducing noise within extended targets while preserving point target, polarimetric signatures and meaningful details. However, optimal speckle filtering requires some knowledge about the underlying target properties. Also, using all the terms of the quadpol polarimetric matrix in order to drive speckle filtering is still an open problem. Deep Learning techniques have been successful in tackling difficult computer vision tasks and show promising results when applied to SAR and PolSAR data. The goal of this paper is to investigate the use of Convolutional Neural Networks (CNN) in order to extract information from the full quadpol covariance matrix and help improve speckle filtering. Experiments are conducted using simulated PolSAR data and results show that we can potentially recover information from the off-diagonal terms.

Earth Observations (EO) enable scientific research, such as the study of meteorology and climate, ecosystems and forests, hydrology and marine life. Applications of EO help protect populations from disasters and improve life in intelligent cities. Increasingly, Machine Learning techniques are seen as key to solve these complex multidisciplinary problems. The scale and dimensionality of data involved often require the definition of processing chains, or workflows. Standards can facilitate the composition, sharing, execution and discovery of these workflows and applications, making them more useful. This paper presents three applications based on Deep Learning: a tree species classifier, a car detector and a flood detector. These applications rely on software containers to package ML framework and algorithms, as well as on workflows to process EO data. We found that these practices allow improved reuse and deployment of research assets in infrastructures. We also note the strong discriminative capabilities of Deep Learning on smaller datasets and the difficulty of gen-eralization to other methods of sensing or regions of interest.

Non-orthogonal multiple access (NOMA) has been developed as a key multi-access technique for 5G. However, secure transmission remains a challenge in NOMA. Especially, the user with weakest channel is most threatened by eavesdropping, due to its highest transmit power. In this paper, we propose a novel scheme to generate artificial jamming at the NOMA base station (BS), aiming at disrupting the potential eavesdropping without affecting the legitimate transmission. In the scheme, the transmit power of artificial jamming is maximized, with its received power at each receiver higher than that of other users. Thus, the jamming signal can be eliminated via successive interference cancellation before others, and the eavesdropping can be disrupted effectively. Due to the non-convexity of the optimization problems, we first convert it to a convex one and then provide an iterative algorithm to solve it. Simulation results are presented to show the effectiveness of the proposed scheme in guaranteeing the security of NOMA networks.

Non-orthogonal multiple access (NOMA) has been developed as a key multi-access technique for 5G. However, secure transmission remains a challenge in NOMA. Especially, the user with weakest channel is most threatened by eavesdropping, due to its highest transmit power. Two schemes are proposed to generate artificial jamming at the NOMA base station (BS), aiming at disrupting the potential eavesdropping without affecting the legitimate transmission. In the first scheme, the transmit power of artificial jamming is maximized, with its received power at each receiver higher than that of other users. Thus, the jamming signal can be eliminated via successive interference cancellation (SIC) before others. When the transmit power of the BS is inadequate, the transmit jamming power is maximized with the jamming signal zero-forced at each receiver. Thus, the legitimate transmission is not affected by the jamming, and the eavesdropping can be disrupted effectively. Due to the non-convexity of these two optimization problems, we first convert them to convex ones and then provide an iterative algorithm to solve them. Simulation results are presented to show the effectiveness of the proposed schemes in guaranteeing the security of NOMA networks.

The ability to send low cost satellites into space has changed the satellite industry and vastly opened up the use of satellites to transform data into information for individuals, organizations, commercial companies, and governments. This information can be corroborated with other sources of data to evaluate the availability of precious resources, e.g., potable water, agriculture; to forecast upcoming famines or diseases; and to perform mapping, communications, and competitive analyses. The cost of owning a satellite is less than $100,000 (using commercial parts) to $1,000 (built by school kids). Launch costs are tens of thousands of dollars, which can be eliminated through subsidies. Compare these costs to $200 million to $1 billion costs for similar functionality, higher performance and greater durability—an interesting trade space that favors multiple low cost flights versus significantly higher costs for permanent satellites for an increasing number of applications. The challenge is to form and manage a development team of unskilled professionals, high school students, or university students to meet deadlines and flight-standards. These team members are likely highly motivated and unskilled. This paper describes the technical management strategy and techniques used to develop the 10 cm, 6-sided CubeSats.

Single Image Super-Resolution (SISR) is looking at restoring the missing high-resolution information from a single low-resolution image in order to increase the apparent spatial resolution by a factor of two or more. In recent years, convolution neural networks have been applied with great success to the problem of improving spatial resolution from a single image. With the advent of low-resolution (10 m) optical sensors such as Sentinel-2, it is interesting to explore the possibility of improving image resolution with Deep Learning (DL) techniques. The purpose of this article is to investigate the potential performances of recent DL super-resolution techniques. The techniques explored here include not only techniques for enhancing high-frequency content but also so-called image-to-image translation techniques based on Generative Adversarial Neural Networks (GAN). From our preliminary results, we show that GANs have the ability to restore complex textural information.

An artificial neural network (ANN)-based method for radio-frequency analog circuit synthesis is implemented on a field-programmable gate array (FPGA). The ANN has four hidden layers, with fifteen neurons per hidden layer, and its hyper parameters are tuned by an auxiliary genetic algorithm (GA) that uses deterministic tournament for generation renewal with minimal hardware. The presented work actualizes the inherently parallel nature of ANN processes, doing away with optimizing vector manipulations by conventional serial hardware. Instead, the effort is put on minimizing the resources used by each neuron and maximizing their collective processing power. Moreover, the GA algorithm for hyper parameter tuning is implemented as a parallel process as well. The proposed architecture is validated on a concrete problem, showing its ability to learn the solution to a problem and generalize it to new instances.

During the design of complex systems implemented in field programmable gate arrays (FPGA), an estimate of about 60% of the designer's time is spent for testing, including nonregression simulations and on-board tests [1]. In this perspective, it is key to guarantee the perfect match of hardware description language (HDL) coding with the algorithm. This has to be achieved without compromise on code quality. To satisfy these requirements, an innovative design methodology has been set-up. We propose a straightforward approach to model the algorithm, in order to ease the implementation task and to reduce turnaround from HDL design back to system modeling. This methodology also enables automation of all steps from the simulation model down to the in-situ test on board. This paper will depict the tools and languages used to co-verify the algorithm with the HDL implementation, down to the delivery. This flow does not use any High Level Synthesis tools, but is rather focused on HDL design productivity improvement. This methodology benefits from open source tools used in software development that we have adapted to the HDL.

Speckle removal from single-channel and multi-dimensional SAR remains a difficult problem. In this paper, we are investigating the use of a Convolutional Neural Network (CNN), previously applied to the Super-Resolution problem, for speckle removal. Because speckle noise statistics is signal dependent, we are training the neural network on the residual image formed by the ratio of the observed intensity to the local mean. Preliminary results on SAR and PolSAR show the potential of CNN to retrieve lost image details.

This paper discusses the Model-Driven Engineering (MDE) of real-time embedded (RTE) systems with soft real-time constraints using UML for Real-Time (UML-RT) and Papyrus-RT. UML-RT is a profile of UML specifically designed for RTE systems. It has a long, successful track record of application and tool support via, e.g., IBM Rational RoseRT, IBM RSA-RTE, and now Papyrus-RT. Papyrus-RT is an Eclipse-based, open-source modelling development environment for UML-RT systems. It allows the generation of complete, executable code from models and advances the state-of-art via support for model representation with mixed graphical/textual notations and an extensible code generator. This paper introduces the central modelling concepts of UML-RT and features of Papyrus-RT. It also presents some advanced features of Papyrus-RT such as import capabilities, mixed graphical/textual modelling, and incremental code generation.

In this paper, we present the design of two smart sensor systems to monitor the gait of patients. These sensor systems were developed for deployment within a body worn wireless network system of medical devices. Telemetry, ambulatory and remote monitoring systems composed of micro-mechanical systems have gained importance in the last decade as medical and rehabilitation institutions try to reduce costs by discharging patients earlier while still requiring various levels of monitoring. Most of the systems currently on the market are bulky, closed architecture, static in configuration and use wired medical devices, all of which limit their usage. Gait monitoring is mainly done in laboratories that are fixed and expensive. The aim of the research which encompasses both systems discussed in this paper is to develop an open architecture using Real-Time Object Oriented Modeling that will allow wireless, wearable medical devices to join a dynamically configurable monitoring environment. The intent of the system is to monitor patients recovery by measuring biometrics and biomedical signals as they go about their daily activities. The sensors that are being developed as part of this research are smart sensors that can provide pre-processed information, reducing the load on the wearable computer.

A cognitive device-to-device (D2D) network with D2D transmitters (DTs) that harvest radio-frequency (RF) energy from the primary transmitters (PTs) is investigated. A novel D2D transmitter-assisted cooperative (DTAC) protocol is proposed, in which a group of DTs that have no transmission opportunity act as potential relays to improve the communications of the primary network. The primary network outage probability is characterized and used to make comparisons between the direct link and the cooperative link which adopts different combining techniques at the primary receivers. The active probability of the DTs is derived, and the D2D network throughput is maximized by seeking an optimal transmission power for the PTs. Simulation results are provided to validate the theoretical analysis.