1,721,001 research outputs found
Fault tolerant motion planning for a quadrotor subject to complete rotor failure
No full textAssuring safety of a quadrotor subject to rotor failure has been heavily investigated at the control level in view of fault tolerant control (FTC) approach. Yet, the existing FTCs are often concerned with tracking the reference motion even when that reference may not be safely trackable due to the physical constraints of the quadrotor. This paper tackles the faulty quadrotor safety at the planner level, proposing a fault tolerant motion planner. Starting from the formal backward reachability problem formulation, the proposed motion planner generates the time trajectory of the coupled rotational and translational motions that safely guide the faulty quadrotor. The generated trajectory is theoretically guaranteed to be tracked by the embedded FTC without violating physical constraints. Further, the trajectory is prescribed as an analytical closed-form expression and thus suitable for real-time emergency maneuvers. The effectiveness of the proposed motion planner is numerically validated in conjunction with the different FTC techniques and compared to the existing planning method. The simulation results clearly signify that the proposed planner can successfully complement the fault tolerance of quadrotor. The supplements including code implementations are available on GitHub repository: https://github .com /HMCL UNIST /Fault-tolerant -motion-planner.
Viewpoint-aware Point Cloud Registration for Robust SLAM
No full textDepartment of Mechanical EngineeringSimultaneous localization and mapping (SLAM) utilizes consecutive sensor inputs to compute its movement over time. During this process, errors continue to accumulate at each step. These accumulated errors are a significant challenge for SLAM. The most well-known solution for these accumulated errors is loop closing. Lidar-based SLAM closes the loop by registering two point clouds. However the point cloud registration is easy to fail when there are missing points between two point clouds due to occlusion and partial overlap. To address this problem, here we proposes a viewpoint-aware point cloud registration method that applies weight in the point-to-plane iterative closest point (ICP) algorithm to consider the visibility of source point cloud from the target point cloud viewpoint. The proposed method yields more accurate and robust results than the baseline approaches in the experiments using KITTI dataset.clos
GPS ????????? ????????? ????????? ????????? ?????? ????????? ????????? ?????? ?????? ??????
No full textDepartment of Mechanical EngineeringCyber Physical System(CPS) integrates physical components such as sensors and actuators and software components such as application and communication software. Most infrastructures that supports our lives such as smart grid, autonomous automobile systems, and robotics systems are typical examples of CPS. Since the physical and software components are strongly interrelated in CPS, a trivial error in software components can result in critical accidents in the physical world. In particular, as signals of sensors and actuators in CPS are becoming more transmitted and received through wireless communication, vulnerability to cyber attacks which could lead to severe incidents is gradually increasing. Among various factors vulnerable to cyber attacks, sensors should be thoroughly protected from cyber attacks since sensor attacks can lead to a serious physical accident, affecting the state estimator and controller directly. In this study, as one of countermeasures against GPS spoofing which is one of the most wellknown sensor attacks, a Secure State Estimation(SSE) is discussed. GPS Spoofing deceives the position and time of GPS receiver by manipulating sensor measurements from satellites or GPS receivers. In particular, the attacks inserted by the attacker in the sensor measurements can be random and are not presumed to follow any specific model. Nonetheless, the developed SSE algorithm makes it possible to accurately estimate the position of GPS receiver even if some of sensor measurements on GPS system are corrupted by an adversarial attacker. In addition, in order to help readers better understand the necessity for SSE, a description of Compressed Sensing(CS), which is the origin of the SSE problem will also
be covered in the preliminaries. Lastly, through the simulation of demonstrating GPS spoofing attack to the GPS receiver of Unmanned Aerial Vehicle(UAV) and defending it with SSE, one can assure the SSE can be promising countermeasure against GPS spoofing attack.ope
???????????? ?????? ????????? ?????? ????????? ??????????????? ?????? ?????? ?????? ?????????
No full textDepartment of Mechanical Engineeringclos
STPA-ML Integrated Fault Injection Test for Autonomous Driving Vehicle
No full textDepartment of Mechanical EngineeringAs autonomous driving system (ADS) technologies are being developed rapidly, the safety and reliability are also being growing concerns on the road. To develop the safe mechanism of autonomous vehicles(AV) against various fault scenarios, comprehensive and exhaustive fault analysis is required a priori. This paper proposes a systems-theoretic processes analysis (STPA) - machine learning (ML) integrated fault injection test (FIT) framework to find meaningful and manageable fault parameters more efficiently and exhaustively against various fault scenarios. First, we design FIT scenarios to reduce the search space of fault parameter by combining STPA and ontology methods. Fault models are selected by STPA which is one of the fault analysis methods and driving scenarios are generated by ontology method that defines the relationships between all objects on the road. By defining fault models and driving scenario, we can reduce the search space of fault parameters at an early stage. Secondly, given FIT scenarios, we design the FIT simulator that contains general ADS and vehicle model to create a driving data set. This data set is labeled for dangerous situations by safety metric. Finally, based on the driving data set, long-short term memory (LSTM) training is implemented to predict the expected time to collision (ETTC). Experimental results indicate trained LSTM predicts the ETTC against various FIT scenarios. This trained LSTM will be used as FIT simulator to expedite simulation time, to cover much more fault scenarios efficiently. Based on the FIT results, we further will organize the corresponding requirements, which can be used in the autonomous driving algorithm to develop the safe ADS as future research.clos
Systematic Fault Injection Scenario Generation for the Safety Monitoring of the Autonomous Vehicle
Get PDFDepartment of Mechanical EngineeringThe Object and Event Detection and Response (OEDR) assessment of Automated Vehicles(AVs) must be thoroughly conducted on the entire Operational Design Domain(ODD) to prevent any potential safety risk caused by corner cases. In response to these challenges, AVs must be tested over hundreds of millions of kilometers before deployment to convince its OEDR capabilities. However, claiming safety through years of testing on the entire ODD is not practically sound. Therefore, many studies have addressed this problem, focusing on efficiently and effectively finding corner cases within high-fidelity simulation environment. In particular, one of key OEDR functionalities is a collision risk assessment system alarming the driver about an impending collision in advance. In AV ODD context, the collision risk assessment is confronting challenging situations such as incorrect sensor information and unexpected algorithmic errors derived from uncertain environments (weather, traffic flow, road conditions, obstacles). Whereas the widely employed collision risk assessment methods relies on the first principle, e.g., Time-To-Collision (TTC), the aforementioned situations cannot be properly assessed without appropriate scene understanding toward the each situation. To this end, AI-based research that leverages previous experience and sensor information (especially camera image) to assess collision risk through visual cues has been developed in recent years. Inspired by the above research trends, this paper aims to develop: 1) systematic corner case generation using a scenario-based falsification simulationand 2) an AI-based safety monitoring system applicable in complex driving scenarios. The implemented simulation is shown to competently find the corner case scenarios, through which the developed system is validated that it can be used as an alternative to an existing collision risk indicator in complex AV driving scenarios.ope
Overtaking in Autonomous Racing with Online Refinement of Opponent Behavior Prediction using Gaussian Process
Get PDFDepartment of Mechanical EngineeringThis paper addresses an overtaking strategy in autonomous head-to-head racing, by virtue of a learningbased prediction to the opponent vehicle???s behavior. The existing prediction approaches either rely on prior model or off-line learning for opponent behavior, whose accuracy diminishes when the opponent in real racing exhibits different driving style. Motivated by this concern, we proposes an online learningbased prediction algorithm that can adapt to the opponents??? different driving style and refine the prediction during the race. Resorting to Gaussian Process (GP) regressor as the baseline learning model, we leverage several techniques to reduce the data size and computation cost of GP, making the algorithm suitable for online learning and prediction refinement in real time. The effectiveness of the proposed algorithm is demonstrated with different simulation scenarios and compared with the other algorithms in terms of prediction accuracy, computation efficiency, and success rate of overtaking maneuver.ope
Cyber attack analysis on cyber-physical systems: Detectability, severity, and attenuation strategy
No full textSecurity of Cyber-Physical Systems (CPS) against malicious cyber attacks is an important yet challenging problem. Since most cyber attacks happen in erratic ways, it is usually intractable to describe and diagnose them systematically. Motivated by such difficulties, this thesis presents a set of theories and algorithms for a cyber-secure architecture of the CPS within the control theoretic perspective. Here, instead of identifying a specific cyber attack model, we are focused on analyzing the system\u27s response during cyber attacks. Firstly, we investigate the detectability of the cyber attacks from the system\u27s behavior under cyber attacks. Specifically, we conduct a study on the vulnerabilities in the CPS\u27s monitoring system against the stealthy cyber attack that is carefully designed to avoid being detected by its detection scheme. After classifying three kinds of cyber attacks according to the attacker\u27s ability to compromise the system, we derive the necessary and sufficient conditions under which such stealthy cyber attacks can be designed to cause the unbounded estimation error while not being detected. Then, the analytical design method of the optimal stealthy cyber attack that maximizes the estimation error is developed. The proposed stealthy cyber attack analysis is demonstrated with illustrative examples on Air Traffic Control (ATC) system and Unmanned Aerial Vehicle (UAV) navigation system applications. Secondly, in an attempt to study the CPSs\u27 vulnerabilities in more detail, we further discuss a methodology to identify potential cyber threats inherent in the given CPSs and quantify the attack severity accordingly. We then develop an analytical algorithm to test the behavior of the CPS under various cyber attack combinations. Compared to a numerical approach, the analytical algorithm enables the prediction of the most effective cyber attack combinations without computing the severity of all possible attack combinations, thereby greatly reducing the computational cost. The proposed algorithm is validated through a linearized longitudinal motion of a UAV example. Finally, we propose an attack attenuation strategy via the controller design for CPSs that are robust to various types of cyber attacks. While the previous studies have investigated a secure control by assuming a specific attack strategy, in this research we propose a hybrid robust control scheme that contains multiple sub-controllers, each matched to a specific type of cyber attacks. Then the system can be adapted to various cyber attacks (including those that are not assumed for sub-controller design) by switching its sub-controllers to achieve the best performance. Then, a method for designing a secure switching logic to counter all possible cyber attacks is proposed and it verifies mathematically the system\u27s performance and stability as well. The performance of the proposed control scheme is demonstrated by an example with the hybrid H2 - H-infinity controller applied to a UAV example
Going Beyond Counting First Authors in Author Co-citation Analysis
Get PDFThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
- …
