[dipl] Perustieteiden korkeakoulu / SCI

Permanent URI for this collectionhttps://aaltodoc.aalto.fi/handle/123456789/21

Browse

Browsing [dipl] Perustieteiden korkeakoulu / SCI by Degree programme/Major subject "Autonomous Systems"

Filter results by typing the first few letters
Now showing 1 - 20 of 23
  • 3D mapping and change detection for patrolling mobile robots
    (2021-08-23) Göncz, Levente
     Perustieteiden korkeakoulu | Master's thesis
    The goal of this thesis was twofold: First, the capabilities of the ZED 2 stereo vision camera were to be assessed to determine whether the camera is suitable to use in real-time applications. Second, the thesis investigated the possibility of implementing a real-time change detection system using the camera. The camera came with very little information regarding its performance. Therefore, in this thesis, a large set of experiments were carried out to evaluate the performance of the camera with respect to depth accuracy and consistency, visual tracking and relocalization accuracy, and object detection performance. Through testing, it has been verified that the camera is able to provide fast and accurate depth measurements, 3D position information and object detection data. Thus, the ZED 2 stereo camera is suitable for real time applications. Furthermore, a 3D change detection system was designed which was built around the concept of semantic object maps using the ZED 2 stereo camera. The proposed algorithm is able to maintain a coherent semantic object map of the environment and detect object-level changes between consecutive patrol routes.
  • Case Study for Semantic Search in Requirements Specification
    (2020-08-18) Asif, Rizwan
     Perustieteiden korkeakoulu | Master's thesis
    Requirements engineering is an integral part of industrial engineering processes, which provides requirements specification in the form of technical documentation. These documents utilize technical natural language which is not very common for other natural language documents. Moreover, tracing or inter-connectivity of requirements is a common practice, which is usually not found in other natural language documents. In this thesis we create a case study to understand requirements engineering practices. The case study is based on creating a search engine that could benefit requirement engineers, while considering the natural language understanding challenge of technical documents. In order to find a better fit for requirement engineers, we instigate with a traditional search engine and then we augment this traditional search engine to train three different models thus creating additional three neural search engines. We used qualitative analysis to assess the effectiveness of each search engine and understand the user needs. This thesis contributes to the natural language understanding of requirements engineering documentation. Our results indicate that plain text frequency based search engines are sufficient for requirements engineers, however, neural models trained with diverse set of data can improve borderline cases and improve the results altogether. These conclusions are limited to qualitative assessment due to lack of comparative data for quantitative assessment.
  • Cooperative Driving Behavior Model for Intelligent Traffic Generation
    (2021-08-23) Triantafyllidis, Theocharis
     Perustieteiden korkeakoulu | Master's thesis
    Intelligent Test Vehicles are following pre-defined trajectory based on the actual scenario. Just the slightest offset from the ideal scenario and trajectory plan might lead to the situation where the desired outcome is unreachable. In order to be able to compensate such delays and offsets a cooperative behavior required between the test vehicles to adapt to new situations and compensate. ITVs are not able to make decisions that would lead to another scenario – their maneuver space is limited. But ITVs should be able to react on the decisions of the Vehicle Under Test (VUT) and they might be able to compensate in order to avoid accidents.
  • Deep Reinforcement Learning for Complete Coverage Path Planning in Unknown Environments
    (2020-12-14) Boufous, Omar
     Perustieteiden korkeakoulu | Master's thesis
    Mobile robots must operate autonomously, often in unknown and unstructured environments. To achieve this objective, a robot must be able to correctly perceive its environment, plan its path, and move around safely, without human supervision. Navigation from an initial position to a target lo- cation has been a challenging problem in robotics. This work examined the particular navigation task requiring complete coverage planning in outdoor environments. A motion planner based on Deep Reinforcement Learning is proposed where a Deep Q-network is trained to learn a control policy to approximate the optimal strategy, using a dynamic map of the environment. In addition to this path planning algorithm, a computer vision system is presented as a way to capture the images of a stereo camera embedded on the robot, detect obstacles and update the workspace map. Simulation results show that the algorithm generalizes well to different types of environments. After multiple sequences of training of the Reinforcement Learning agent, the virtual mobile robot is able to cover the whole space with a coverage rate of over 80% on average, starting from a varying initial position, while avoiding obstacles by using relying on local sensory information. The experiments also demonstrate that the DQN agent was able to better perform the coverage when compared to a human.
  • Design and development of a time-synchronized positioning platform using FPGA
    (2020-08-18) Sléber, Botond
     Perustieteiden korkeakoulu | Master's thesis
    The positioning platform is an extended Integrated Navigation System (INS), where the data from the Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU) are combined with additional suitable and complementary sensors, whereas the platform is built from commercial off-the-shelf (COTS) components. For extending the INS capabilities, altimeter and magnetometer is integrated in the system as well. The research is focusing on finding an optimal architecture for synchronizing the sensory data, combining them with the appropriate timestamps. Proper, reasonable and robust time-synchronization scheme is designed for obtaining the precise event times. The development implements it in an FPGA SoC with the aim of fast and reliable data processing and management, under 1 us synchronization error. The implementation within the FPGA consists of the development of the communication with the serial interfaces and the implementation of the timestamping and time synchronization functionalities create the complete data frames. The system design involves a host device, where the actual signal processing algorithms are supposed to be running in later stages of the project and maintains two-way communication with the FPGA. This thesis describes the design and development of a positioning platform, consisting GNSS and INS with the aim of integrating them into one platform, while processing the measurements into one, accurate and synchronized time domain.
  • Detecting obstacles from camera image at open sea
    (2020-08-18) Knébel, Miklós
     Perustieteiden korkeakoulu | Master's thesis
    While self-driving cars are a hot topic in these days, fewer people know that the same level of automation is being developed in the maritime industry. To enhance safety on board and to ensure the optimal utilization of crew members, automated assistant solutions are implemented on cargo ships and vessels. This thesis deals with a monocular camera-based system, that is capable of detection obstacles in open sea scenarios, and to estimate surrounding vehicles’ distance and bearing. After a solid research of existing methods and literature, an algorithm has been developed, containing three main parts. First, the real-world measurement data and camera images are being processed. Secondly, object detection is achieved with the YOLO deep learning methods that achieves at a high framerate and can be used for real-time applications. Lastly, distance and bearing values of detected obstacles are estimated based on geometrical calculations and mathematical equations that are validated with ground truth measurement data. Having multiple weeks of recorded measurement data from a RoPax vessel operating from Helsinki, allowed testing and validation already during the development phase. Results have shown that the systems’ detection capability is highly affected by the image resolution, and that distance estimation performance is reliable until 2-3 kilometers, but estimation errors rise at farther distances, due to physical limitations of cameras. In addition, as an interesting evaluation method, a survey has been conducted with industry professionals, to compare human distance estimation capability with the developed system. As a conclusion it can be stated that a significant need and huge potential can be found in automated safety solution in the maritime industry.
  • Evaluating the utility of multi-user VR in product development
    (2020-08-18) Gabriel, Alberto
     Perustieteiden korkeakoulu | Master's thesis
    Verifying the design of a mechanical product typically requires a physical mockup. Virtual reality can avoid this by letting workers experience real-size design inspections avoiding to waste time and money to produce physical and potentially still faulty assets. Multi-user environments have been studied to verify the utility of virtual reality in performing design reviews for maintainability. The main features required in such software have been determined. A systematic literature review on the state of the art usages of virtual reality for design inspections and industrial training completes the thesis.
  • Feasible and adaptive attention-based models for multimodal trajectory prediction in urban driving scenarios
    (2020-10-20) Franceschini, Riccardo
     Perustieteiden korkeakoulu | Master's thesis
    A self-driving car which takes an autonomous decision needs three main building blocks, a perception module, a prediction module and a planning module. In this thesis, we consider the vehicle already capable of understanding the surrounding area; thus, we focus on the prediction module, which is responsible for predicting the future of the other agents in the scene. Thus, we examine, in particular, the prediction in urban driving scenarios in a multimodality setting where the model can learn to predict all the possible future scenarios in such complex environment. The predictions consist then of multiple sequences of coordinates plus the probability for each future. After having investigated the past and current methods, we have implemented different baselines, both deep learning methods and not. Hence, examining both the data that we used, and the network structures, we believe that some improvements are possible, and here we propose some methods to address those problems. First, we extend the previous loss with an additional term called offroad loss that penalise the model when the prediction lays outside of the road structure. Second, considering also that difficult scenes are rarer than simple scenes, we propose two different weighted sampling methods to overcome such imbalance, in this way, the model can adapt the prediction to more complicated and rare scenes. Finally, we try to extract more useful information from road structure, nearby agents and past information implementing different attention architectures inside the models. In this thesis, we also conduct a performance comparison between our methods and the baselines applying commonly used metrics. Moreover, to visually understand the impact of each method, we propose some anecdotal analysis showing the real differences in terms of prediction in some challenging situations.
  • Improving Ad-Hoc Cooperation in Multiagent Reinforcement Learning via Skill Modeling
    (2020-08-18) Kwiatkowski, Ariel
     Perustieteiden korkeakoulu | Master's thesis
    Machine learning is a versatile tool allowing for, among other things, training intelligent agents capable of autonomously acting in their environments. In particular, Multiagent Reinforcement Learning has made tremendous progress enabling such agents to interact with one another in an effective manner. One of the challenges that this field is still facing, however, is the problem of ad-hoc cooperation, or cooperation with agents that have not been previously encountered. This thesis explores one possible approach to tackle this issue, using the psychology-inspired idea of Theory of Mind. Specifically, a component designed to explicitly model the skill level of the other agent is included, to allow the primary agent to better choose its actions. The results show that this approach does in fact facilitate better coordination in an environment designed to test this skill and is a promising method for more complicated scenarios. The potential applications can be found in any situation that requires coordination between multiple intelligent agents (which may also include humans), such as traffic coordination between autonomous vehicles, or rescue operations where autonomous agents and humans have to work together to efficiently search an area.
  • Indoor Localization of AGVs: A Ground Texture Based Solution
    (2020-12-14) Macías Solá, Javier
     Perustieteiden korkeakoulu | Master's thesis
    In the era of digitalisation of industrial processes, autonomous indoor robots are gathering increasing attention due to their maturity and the advantages they can bring into existing factories. These can boost productivity while reducing costs of relatively simple tasks such as moving goods within the stations of a warehouse. All these intralogistic operations must be performed under control and with very high precision. However, existing popular solutions require factory owners to adapt their infrastructure to fit the requirements of the AGVs. These solutions usually include the addition of artifical landmarks that the robot can detect with a dedicated sensor, as it is the case of placing lines or QR codes on the floor. Although these solutions allow the robot to globally localise itself with millimeter accuracy, they require an initial investment that is not necessarily free of issues. For instance, lines can be damaged by heavy machinery and have to be replaced recurrently. This master’s thesis aims at providing an alternative solution to indoor localisation of AGVs that does not rely in any artificial landmark that is not already available in the average factory. After analysing existing solutions for a wide variety of sensor principles as well as our requirements, we find ground texture based localisation to be a competitive candidate for this task. Our implementation adapts some ideas expressed in Identity Matching [1] and Micro-GPS [2] to create a fundamentally different localisation method using ground textures. Although the implemented solution is still far from production, this technology complies with all the business requirements. In our evaluation on a custom dataset recorded at various industrial sites, we find the computational time to be a challenging part, as well as occlusion reducing the localisation success rate considerably.
  • Innovation Activity CityBeamer
    (2020-10-20) Wissing, Gijs
     Perustieteiden korkeakoulu | Master's thesis
  • Interpreting Multivariate Time Series for an Organization Health Platform
    (2021-03-15) Saluja, Rohit
     Perustieteiden korkeakoulu | Master's thesis
    Machine learning-based systems are rapidly becoming popular because it has been realized that machines are more efficient and effective than humans at performing certain tasks. Although machine learning algorithms are extremely popular, they are also very literal and undeviating. This has led to a huge research surge in the field of interpretability in machine learning to ensure that machine learning models are reliable, fair, and can be held liable for their decision-making process. Moreover, in most real-world problems just making predictions using machine learning algorithms only solves the problem partially. Time series is one of the most popular and important data types because of its dominant presence in the fields of business, economics, and engineering. Despite this, interpretability in time series is still relatively unexplored as compared to tabular, text, and image data. With the growing research in the field of interpretability in machine learning, there is also a pressing need to be able to quantify the quality of explanations produced after interpreting machine learning models. Due to this reason, evaluation of interpretability is extremely important. The evaluation of interpretability for models built on time series seems completely unexplored in research circles. This thesis work focused on achieving and evaluating model agnostic interpretability in a time series forecasting problem. The use case discussed in this thesis work focused on finding a solution to a problem faced by a digital consultancy company. The digital consultancy wants to take a data-driven approach to understand the effect of various sales related activities in the company on the sales deals closed by the company. The solution involved framing the problem as a time series forecasting problem to predict the sales deals and interpreting the underlying forecasting model. The interpretability was achieved using two novel model agnostic interpretability techniques, Local interpretable model- agnostic explanations (LIME) and Shapley additive explanations (SHAP). The explanations produced after achieving interpretability were evaluated using human evaluation of interpretability. The results of the human evaluation studies clearly indicate that the explanations produced by LIME and SHAP greatly helped lay humans in understanding the predictions made by the machine learning model. The human evaluation study results also indicated that LIME and SHAP explanations were almost equally understandable with LIME performing better but with a very small margin. The work done during this project can easily be extended to any time series forecasting or classification scenario for achieving and evaluating interpretability. Furthermore, this work can offer a very good framework for achieving and evaluating interpretability in any machine learning-based regression or classification problem.
  • Machine Recognition of Engineering Diagrams in Process Industry
    (2020-10-20) Qu, Rui
     Perustieteiden korkeakoulu | Master's thesis
    Engineering diagrams are widely used in process industry as a standard graphic language to represent engineering schematics and convey information. Over the years, a large amount of legacy engineering diagrams has been accumulated in companies so that there is an increasing demand on digitizing the diagrams to improve productivity. With the progress of computer vision, especially deep learning-based object detection, we take advantage of the latest deep learning models and algorithms to process and recognize the legacy diagrams, which fuel the networks. In this thesis, an end-to-end digitizing model is proposed to recognize engineering diagrams as machine encoded format. Due to the complexity of diagrams, the recognition task is divided into three sub-targets: symbols, connectivity and characters. We experiment on multiple state-of-the-art deep learning-based approaches to recognize symbols, such You Only Look Once (YOLO). Line recognition algorithm is proposed based on Hough transformation and Skeletonization. For the characters, we conduct the recognition by two steps, first locating, second recognizing. The model is evaluated on real industry engineering diagrams with quantitative and visual results provided. Firstly, the results demonstrate that YOLO works well for symbol recognition, reaching over 90% mAP@0.75 of all symbols. YOLO can also be used for character locating, where the characters are regarded as a symbol. Secondly, connection lines can be recognized effectively by the proposed algorithm combining Hough Transformation with region of interests. The result of recognizing the three sub-targets are integrated to generate a DXF format diagram. There is also some discussion on a universal model which can expand the usage of our model to different types of diagrams. One of the most essential steps is the analysis of source diagrams and data preparation, which is time consuming dirty work but can significantly improve the recognition performance.
  • Modeling and detection of cable collisions for collaborative Cable-Driven Parallel Robots
    (2021-12-13) Rousseau, Thomas
     Perustieteiden korkeakoulu | Master's thesis
    This Master’s Thesis studies the problem of cable collisions with the environment for Cable-Driven Parallel Robots (CDPR) in the context of collaborative robotics. When operators are working in the same workspace as a CDPR, they are very likely to be in contact with cables at some point because these span a large portion of the workspace. Hence, this work aims at defining a way to tolerate light collisions so as to maintain a continuous workflow in a safe manner by proposing a collision detection method. In case dangerous collisions are detected, appropriate actions to mitigate their impact need to be defined. An overview of the work present in the literature regarding CDPR safety is first presented, focusing on cable collisions and post-failure recovery. Then, collisions between the cables of a Cable-Driven Parallel Robot (CDPR) and the environment are modeled using a simple punctual contact model with a lateral force. This model returns a tension increase inside the collided cable that can be determined knowing its initial tension and the collision force. The lateral collision force can be bounded by values excerpted from international safety standards, allowing to define thresholds for the tension increase in the cables due to a collision. This model is then validated experimentally using an adapted test bench. Then, by monitoring the tension in the cables, it is possible to determine whether the robot dangerously collided with the environment if a cable saw its tension increase over the admissible threshold determined with the collision model. An operational implementation of this detection method on a CDPR prototype is introduced and evaluated for two different collision types. A safe mitigation strategy is also proposed to limit the damage that could result of such dangerous collisions, relying on finding alternative tension distributions able to maintain the equilibrium of the platform while releasing an over-taut collided cable if the robot pose allows it. Several possibilities are also proposed if the alternative equilibrium cannot be found on the current pose, relying on previous work in the literature conducted on cable failures.
  • OPC UA Integration within the CERN Control and Monitoring Framework
    (2020-10-20) Stockinger, Elisabeth
     Perustieteiden korkeakoulu | Master's thesis
    The open-source CERN Control and Monitoring framework (C2MON) was designed to consolidate information from the technologically and contextually diverse monitoring entities operating within the context of CERN's particle acceleration experiments. Its support for OPC UA, a widely successful and open industrial communication protocol, has become outdated, and it cannot interface with modern generations of OPC UA servers. This thesis analyzes current and foreseeable user needs within the context of CERN. Based on design science research methodology, a modern OPC UA data acquisition module is designed, implemented, and evaluated according to the elicited user needs. The most impactful modernizations are a framework for extending the configuration model within the given project structure, and support for the OPC UA redundancy model. The subsequent evaluation is centered on the quality requirements of functionality and reliability. It combines an extensive unit and integration test framework with an evaluation of tolerance to faulty network conditions. The potential for innovation forged by the new module was analyzed through expert interviews. Some supervised systems currently expose alarms to C2MON via JMS, but are capable of exposing data points directly through OPC UA instead. Such a shift is enabled by the new module. It reduces architectural complexity and allows the definition of alarms directly through the standard C2MON procedures. This unlocks previously unavailable functionality in terms of configuration verification. The efficacy of the OPC UA module as an interface to a simulated plant was verified in a proof of concept deployment with power supply simulation engines. This allows the investigation of the high-level behavior and reliability of C2MON under conditions exceeding those possible in production and laboratory testing. Given OPC UA's relevance in industrial innovation and as an integration platform, this module ensures that C2MON remains cutting edge and enables the creative adaption of OPC UA based features.
  • Point Cloud Data Augmentation for 4D Panoptic Segmentation
    (2023-01-23) Jin, Wangkang
     Perustieteiden korkeakoulu | Master's thesis
  • Point Pair Feature-based 6D pose estimation using depth maps
    (2020-08-18) Saljoughi, Nicola
     Perustieteiden korkeakoulu | Master's thesis
    Vision-based six degrees-of-freedom (6D) pose estimation of rigid objects is a fundamental task for the development of autonomous systems, playing an important role in robotic grasping, object inspection and manipulation. Model-based methods that rely on Point Pair Features (PPFs) currently outperform learning-based methods in terms of recall on the popular BOP benchmark. In this thesis we thoroughly review model-based methods aiming at identifying limitations and understanding future research directions. We achieve this by implementing the original method of Point Pair Features, by analysing the issues that affect its performance on the Linemod dataset of BOP, and finally by modifying some modules based on the identified limitations. Our improvements allowed us to improve recall from 14% to 36%. Our conclusion is that PPF-based methods are highly reliant on the quality of normal information, thus sensitive to sensor noise. Future research should focus on making the methods less reliant on normal information, for example defining new features by using triplets of points instead of point pairs.
  • Radar-based ego-motion estimation
    (2024-01-22) Mutti, Giacomo
     Perustieteiden korkeakoulu | Master's thesis
    With the increasing levels of automation in driver assistance systems, the expectations for the performance and reliability of automotive Radio Detection and Ranging (RADAR) sensors have grown considerably. Given that a substantial part of RADAR signal processing relies on signal history, precise compensation for vehicle displacement, or "ego-motion", in RADAR signals is crucial for maintaining a coherent signal history. Currently, this compensation is dependent on ego-motion information sourced from the vehicle bus, linking the radar performance to the quality and delay of signals from the vehicle. To mitigate this dependency, it is advantageous to directly estimate vehicle displacement from RADAR sensor data. This thesis introduces an innovative approach to ego-motion estimation leveraging the sensor’s 4D measurement capabilities, incorporating quality indicators to provide reliability feedback for each estimation. This is a pivotal requirement for Advanced Driver Assistance Systems (ADAS) applications, ensuring that every output value is accompanied by an indicator conveying the level of trust it holds for subsequent applications relying on it. Various error indicators gauge the algorithm’s performance, with the estimated egomotion demonstrating high accuracy. These values are complemented by indicators proven to accurately reflect their reliability.
  • Semantic scene understanding and traversability estimation for off-road vehicles
    (2021-08-23) Copado Rodriguez, Jesus
     Perustieteiden korkeakoulu | Master's thesis
  • Structured light assisted real-time stereo photogrammetry for robotics and automation. Novel implementation of stereo matching
    (2020-08-18) Losi, Jacopo
     Perustieteiden korkeakoulu | Master's thesis
    In this Master’s thesis project a novel implementation of a stereo matching based method is proposed. Moreover, an exhaustive analysis of the state-of-the-art algorithms in that field is outlined. Specifically, both standard and deep learning based methods have been extensively investigated, thus to provide useful insights for the designed implementation. Regarding the developed work, it is basically structured in the following manner. At first a research phase has been carried out, hence to simply and rapidly test the thought strategy. Subsequently, a first implementation of the algorithm has been designed and tested using data available from the Middlebury 2014 dataset, which is one of the most exploited dataset in the computer vision area. At this stage, numerous tests have been completed and consequently various changes to the algorithm pipeline have been made, in order to improve the final result. Finally, after that exhaustive researching phase the actual method has been designed and tested using real environment images obtained from the stereo device developed by the company, in which this work has been produced. Fundamental element of the project is indeed that stereo device. As a matter of fact, the designed algorithm in based on the data produced by the cameras that constitute it. Specifically, the main function of the system designed by LaDiMo is to make the built stereo matching based procedure simultaneously faster and accurate. As a matter of fact one of the main prerogative of the project was to create an algorithm that has to prove potential real-time results. This has been in fact, achieved by applying one of the two methods created. Specifically, it is a lightweight implementation, which strongly exploits the information coming from the LaDiMo device, thus to provide accurate results, keeping the computational time short. At the end of this Master’s thesis images showing the main outcomes obtained are proposed. Moreover, a discussion regarding the further improvements that are going to be added to the project is stated. In fact, the method implemented, being not optimized only demonstrate a potential real-time implementation, which would be certainly achieved through an efficient refactoring of the main pipeline.