previous day
all days

View: session overviewtalk overview

09:00-09:40 Session 6: Invited speaker

Philippe Fillatreau: Virtual Reality and interactive and immersive planning for the assistance to manipulation or navigation


Location: TK3
Virtual Reality and interactive and immersive planning for the assistance to manipulation or navigation

ABSTRACT. In industry, whereas the economic competition increases, up-to-date industrial products are more and more integrated and the tasks related to their lifecycle (assembly, maintenance, disassembly…) have to be performed under sometimes very strong geometric constraints. In the context of Industry 4.0 (Factory of the Future) and PLM (Product Lifecycle Management), industrial companies therefore express the needs to validate these tasks from design stage on, in order to be able to update the design of these products (before manufacturing the physical prototypes) if needed. Such an approach allows to reduce development time and cost, to detect errors as early as possible, and to target more environment friendly development processes. When simulating such complex scenarios, it is necessary to deal with the relative positioning or to the movement of objects and of resources (machines, robots, human operators) that manipulate them. A key issue is then to find a path, a trajectory, a movement to show the feasibility of scenarios and simulate what the execution of a task will be. Our works deal with the assistance to the simulation and validation of such complex scenarios in Virtual Reality. We present the original scientific approach on which these works are based: the joint use of motion planning and VR techniques to validate the feasibility of the movement for the simulated scenarios in an interactive and immersive way, with visuo-haptic guidance to the human operator in the loop. The initial approach was based on the use of purely geometric models. In order to improve the relevance of the assistance and the modalities of interaction and control sharing (authority sharing and intents detection) between the VR platform and the human operator, we then considered higher abstraction level (topological and semantic) data than the purely geometrical data traditionally used. Finally, for a better, task- or trade-oriented assistance, and in order to move from the "virtual experience" to the "trade-oriented experience", our work now targets the development of joint, interactive and immersive task and path planning strategies.

09:40-10:40 Session 7: Robotics


Location: TK3
Path planning control using high abstraction level environment model and industrial task-oriented knowledge

ABSTRACT. In order to face an increasing economic competition, industrial manufacturers wish to reduce the time and cost of product development. Furthermore, up-to-date products are more and more integrated, and have to be assembled, disassembled or maintained under potentially very strong geometric constraints. In the context of Industry 4.0, manufacturers are therefore expressing the desire to validate all the tasks related to their products lifecycles, from design stage on, by simulation using a digital mock-up, and before building the physical prototypes. A key issue is then to find a trajectory, a movement, to show the feasibility of the simulated scenarios. Automatic path planning algorithms, developed by the robotics community from the 1980s on, have been widely used for this purpose. In this paper, we intend to improve the relevance of the trajectories proposed by such algorithms and the associated computation times. In order to do so, we consider: a) the use of path planning algorithms or of combinations of these; b) the involvement for the environment modelling of data with a higher abstraction level than the purely geometric data traditionally used [Cailhol et al., 2019]; and c) the representation of the knowledge related to the task to be performed by using ontologies [Zhao et al., 2018]. The approaches developed and associated improvements of the state of the art are validated experimentally through the simulation of highly geometrically constrained manipulation tasks.

Generative design of 3D printed grippers for robot/human colaborative environments

ABSTRACT. The aim of this paper is to design a gripper for disassembling the refrigerator door gasket with a collaborative robot. The design concept is based on a strong and fixed finger used for guiding, whereas the second finger moves to pinch the gasket. The working conditions of the gripper have been characterised in a universal testing machine, and results have shown that the vertical force for extracting the gasket is lower than the nominal 140 N force of the collaborative robot, taking into consideration that its maximum displacement speed is 250 mm/s. The second input for the generative design based on topology optimisation is the design space, with an asymmetrical distribution for the moving and static finger. The resulting optimised material distribution has been reinterpreted taking into consideration the design for additive manufacturing principals. Finally, the gripper has been 3D printed with a short carbon fibre reinforced polyamide in a filament extrusion machine. The resulting gripper is a 40% lighter than the monolithic solutions designed and manufactured by conventional technologies. .

Leveraging domain specific modeling to increase accessibility of robot programming

ABSTRACT. Despite the popularity of the Robot Operating System (ROS), Model-Driven Engineering (MDE) methods remain not commonly used in the community. The use of such methods would allow to reason at a higher level of abstraction and increase the accessibility of field engineers to effectively write ROS applications, while opening the way to more advanced model analysis for validation and verification. In a previous work, we made an in-depth comparative study to choose which MDE approach to apply when developing ROS applications. This study led us to choose graphical Domain-Specific Languages (DSL) as the means to facilitate the development of ROS applications. In this paper we present the MDE4ROS framework, a graphical DSL we defined to allow high level robotics applications development. This framework aims to provide a better view of the system during the development using a graphical representation and to enable higher level of abstraction with code generation of the ROS system. To illustrate our proposal we use MDE4ROS to develop a TurtleBot use case and we present an analysis of the results.

Cetratus: Live Updates in Programmable Logic Controllers

ABSTRACT. Manufacturing companies are facing new market demands, mostly driven by global competition and digitalization. In this context, more efficient, flexible, adaptable and evolvable mechatronic and manufacturing systems are required, which enable quick adjustments to the production in order to address (all these) market changes. However, production idle times due to such re-configurations and adaptations might be costly. In this paper, a live updates concept for Programmable Logic Controllers (PLCs) is presented. The proposed design employs a Petri net runtime engine, in which the executed functional program (the Petri net model with its interpretation) is updated while running, without system shutdown and restart being needed. To this end, a quarantine-mode execution and monitoring approach is used for the new PLC program functional validation. A reconfigurable Vernadat machine case study is also presented.

10:40-11:00Coffee Break
11:00-12:00 Session 8: Information, image and signal processing

Power electronics

Location: TK3
Audio-Visual Broadcast Transcription System Using Artificial Neural Networks

ABSTRACT. In this paper, a new system for audio and visual TV-broadcast News transcription is proposed. In the last few years, our system for audio-only broadcast transcription has been modified with the possibility of obtaining additional visual information, especially from TV video recordings. New extension modules and algorithms mainly for visual information extraction are described in this contribution. Combined Deep Neural Networks with Hidden Markov Models (DNN-HMM) are used for audio speech signal recognition. A classification of a relevant visual signal was based on Convolutional Neural Networks (CNN). There are the additional modules for detection and identification of human faces, TV logos, and company logos in the newly developed transcription system. Another module was designed for Optical Character Recognition (OCR) of text, which occurs mainly in video recordings of TV-News very often. The whole audio-visual system for broadcast transcription was tested on a relatively big database (817 hours) which has been completely transcribed. The system also includes the possibility of intelligent search in transcribed data from audio and/or visual signal.

A State Space Solution to the Estimation of Interacting Vehicle Trajectories with Deep Neural Networks and Variational Bayes Filtering
PRESENTER: Tristan Klempka

ABSTRACT. This paper addresses the estimation of trajectories of interacting vehicles at a microscopic scale, as a prerequisite to their prediction for risk assessment. A state space solution is investigated, where both the Markov hidden state (continuous-valued, which captures the joint histories of vehicles) and the measurements (low-dimensional and noisy) admit a vehicle-wise structure. The vehicles' transition models are assumed independent of each other, time- and vehicle-invariant, and coequal to an "egocentric" prior dynamics pdf. To cope with the vehicles' interactions, this pdf is conditioned on the full state vector as the past time index, what imposes a centralized estimation/prediction of the fleet motion. The two fundamental pillars of the approach are developed: learning of a Gaussian mixture egocentric transition model by means of Deep Neural Networks; synthesis of a stochastic variational Bayes filtering algorithm which features a decentralized vehicle-wise structure but takes into account interactions. Tests on highway scenarios are presented.

A-posteriori synchronization of power quality data in Smart grids

ABSTRACT. This paper is concerned with synchronizing smart meter data in post-processing. A-posteriori synchronization is necessary in cases where real time online synchronization failed or was unavailable. In the paper, a simple method to synchronize power quality data series from multiple meters is proposed. The method analyses similarities in voltage data and estimates the lags between the meters. In order to obtain a single data series representing the most relevant phenomena that happen at the location of the smart meter, Singular Value Decomposition is used. It helps to decrease the dimensionality of the smart meter data to one. To synchronize the time of two or more metering devices the correlation between data series were calculated. Distance between zero and the maximum absolute peak of the cross-correlation is taken as a lag. For signals containing transients and dips, accuracy may be increased by correlating the first order difference of the signals. The output of the proposed algorithm is the value of signals' time lags. The accuracy of the algorithm is limited by the aggregation level of data under examination.

Image Enhancement using GANs for Monocular Visual Odometry

ABSTRACT. Drones, mobile robots, and autonomous vehicles use Visual Odometry (VO) to move around complex environments. ORB-SLAM or deep learning-based approaches like DF-VO are two of the state-of-the-art technics for monocular VO. Those two technics perform correctly in outdoor scenarios but show some limitations in indoor environments. The extreme lighting conditions, non-Lambertian surfaces, or occlusion of indoor environments can disturb the visual information, and so the odometry information. Generative Adversarial Network(GAN) architectures recently proposed in the literature can help overcome image low-light and blurring limitations. This research study aims to assess image enhancement’s impact using GANS on the Visual Odometry algorithm DF-VO. Since DF-VO is also based on visual geometric information, the paper first considers the effect of three different GAN architectures in the camera’s calibration. Then, the impact in the odometry information computed by DF-VO is evaluated. The preliminary results show that the reprojection error and the uncertainty of the calibration of a pin-hole-based camera do not increase significantly, and DF-VO’s performance is improved. Index Terms—Image enhancement, Calibration, Visual Odometry, Deep Learning

12:00-14:00Lunch Break
14:00-14:30 Session 9: Integrated circuits / Smart systems

Electrical machines

Location: TK3
Indoor environment monitoring as a measure to reduce epidemic spreading

ABSTRACT. The Covid-19 pandemic has highlighted a major problem that has been neglected for a long time for various reasons - proper ventilation of indoor spaces in public buildings, schools, the industry, but also in households. The national legislation of developed countries usually sets some ventilation requirements, but their parameters often differ and, most importantly, they are practically unenforceable in everyday life. According to WHO, the majority of the population of European countries spend up to 90% of their time indoors, which by the way significantly contributed to the massive spread of the disease, transmitted mainly by inhalation of aerosols containing the Covid-19 virus. Our paper aims to describe the current state of legislation on indoor environment quality, with some studies dealing with population behavior and its impact on their health. Another goal of our paper is to describe a method that can detect the air exchange quality in a building and predict development of the indoor environment in order to be able to take appropriate measures well in advance either by technical means of forced ventilation or by conventional methods of manual ventilation. In the future, timely and adequate ventilation should not only be one of the important ways to reduce spreading of respiratory diseases in the population but also, more generally, an important way to increase the quality of life and health of the European population.

Design and Performance Analysis of Energy Efficient 11T SRAM (E2S11T) Cell for High Performance and Low Power Applications
PRESENTER: Sargunam Tg

ABSTRACT. The SRAM based cache memory has been radically increasing in low power applications. The on-chip data processing and computations have been consistently increasing for Wireless Sensor Networks (WSN) and Internet-of-Things (IoT) applications. This demands the constant improvement over the power, performance, stability, and energy efficiency. The challenges rely on power and performance of SRAM as the technology node reduces. In this paper, design of Energy Efficient SRAM (E2S11T) cell is proposed. The proposed cell contains 11- Transistors and implemented using 45 nm CMOS technology. The average dynamic power of the proposed cell is minimized by of 86.68%, 86.77%, 61.48% and 38.47% compared to C6T, S8T, LPHS10T and HSF11T cells, respectively. The write delay is reasonably improved about 32.58%, 34.36% and 8.43% against C6T, S8T and LPHS10T cells respectively. The read delay is also improved as well as stability have been improved due to three transistors. The proposed E2S11T cell is proven to be stable in worse conditions against temperature and works without any degradation as low as 200 mV. The cell is statistically analysed by performing Monte-Carlo (MC) simulation to validate the stability of the cell.