Real-time modeling of the surrounding environment is a key functionality for autonomous navigation. Bird view grid-based approaches have interesting advantages compared to feature-based ones. Methods able to encode oc...
详细信息
ISBN:
(纸本)9781509018901
Real-time modeling of the surrounding environment is a key functionality for autonomous navigation. Bird view grid-based approaches have interesting advantages compared to feature-based ones. Methods able to encode occupancy information and to manage perception uncertainty in dynamic environments are quite well known but very few studies have been carried out on encoding semantic information in grids. this kind of information can be crucial in many situations in order to make the vehicle able to follow basic road rules, such as lane keeping or lane changes. Usual approaches often detect lane markings using on-board cameras or lidars but the problem is tricky when the road is multi-lane or in challenging weather conditions. In this work, we propose to tackle this problem by using a vectorial prior map that stores detailed lane level information. From a given pose estimate provided by a localization system, we propose an evidential model that encodes lane information into grids by propagating the pose uncertainty on every cell. this evidential model is compared with a classical Bayesian one and some of its special characteristics are highlighted. Real results carried on public roads withthe real-time software are reported to support the comparison.
One of the most important problems towards studying complicated networks is community detection. the methodology of non-negative matrix factorization (NMF) has lately emerged as among the hottest research issues withi...
详细信息
ISBN:
(纸本)9781665493901
One of the most important problems towards studying complicated networks is community detection. the methodology of non-negative matrix factorization (NMF) has lately emerged as among the hottest research issues within community detection because of its ability to reveal natural structures and trends in high-dimensional information. the primary difficulty is that most community detection methods are hampered by the issue of nodes belonging to several communities. We will use the NMF technique in this work to tackle this issue by creating a novel mathematical function. In addition, we will include a regularized factor for simulating latent embedding space and a correlation factor to prevent overlap inside nodes that belong to various communities. Following that, the entire objective function will employ an optimization approach to arrive at the variable values that are ideal. Finally, we assess the effectiveness of different methodologies on real networks. According to experimental findings, the introduced approach is better among the other state-of-the-art methodology.
One of the biggest challenges towards fully automated driving is achieving robustness. Autonomous vehicles will have to fully recognize their environment even in harsh weather conditions. Additionally, they have to be...
详细信息
ISBN:
(纸本)9781509018901
One of the biggest challenges towards fully automated driving is achieving robustness. Autonomous vehicles will have to fully recognize their environment even in harsh weather conditions. Additionally, they have to be able to detect sensor and algorithm failures and react properly to keep the vehicle in a safe state. these two challenges are addressed exemplarily on miniature cars. We extend the approach of Compositional Hierarchical Models [1] by temporal fusion to achieve a robust environment perception. the increased association problem is overcome by a grid-based approximation and a voting system. System performance assessment surveils the system's performance and reacts with driving function degradation or activation of specialized algorithms. the approach was evaluated at the final of the Audi Autonomous Driving Cup 2016. A video shows the advanced driving capabilities under harsh environment conditions and the source code is available for download.
Facing various uncertainties in real world traffic, navigation services are typically formulated as a certain stochastic shortest path problem (SSPP). In the past several years, many stochastic objectives in SSPP has ...
详细信息
ISBN:
(纸本)9781509018901
Facing various uncertainties in real world traffic, navigation services are typically formulated as a certain stochastic shortest path problem (SSPP). In the past several years, many stochastic objectives in SSPP has been proposed, among which maximizing the probability of arrival on time draws certain attentions from researchers as it takes, besides absolute travel time, an extra dimension of information (user-specified deadlines) into consideration. this paper extends a recently proposed data-driven approach for SSPP in the following two aspects: (1) when deadline is loose enough, the original method may return more than one paths since all of them meet deadline with 100% probability. We propose to return a unique path withthe least number of intersections by reformulating the objective function. (2) we extend the arrive-on-time problem to the case of visiting several fixed locations, which is a common scenario in real world application. Experimental results show the accuracy and effectiveness of the improved data-driven method.
this paper presents an algorithm, called the Backwards Incremental System Optimum Search (BISOS) for achieving system near-optimum traffic assignment by incrementally limiting accessibility of roads for a chosen set o...
详细信息
ISBN:
(纸本)9781509018901
this paper presents an algorithm, called the Backwards Incremental System Optimum Search (BISOS) for achieving system near-optimum traffic assignment by incrementally limiting accessibility of roads for a chosen set of agents. the described algorithm redistributes traffic volumes homogeneously around the city and converges significantly faster than existing methods for system optimum computation in current literature. Furthermore, as previous methods have mainly been developed for theoretical purposes, the solutions provided by them do not contain all the necessary information for a practical implementation such as explicit paths for the commuting population. In contrast, the BISOS algorithm preserves the information about the exact paths of all commuters, throughout the whole process of computing the system optimum assignment. Furthermore, a realistic traffic scenario is simulated using Singapore as a case study by utilizing survey and GPS traffic data. the BISOS routing method needs 15 times less routing computations to get within 1% of the optimal solution for a simulated scenario compared to conventional methods for system optimum computation.
To alleviate the workload of labeling before estimating certain color distributions, integrative labeling is introduced, which merely needs to figure out whether a picture contains positive-class regions or not and th...
详细信息
To alleviate the workload of labeling before estimating certain color distributions, integrative labeling is introduced, which merely needs to figure out whether a picture contains positive-class regions or not and then all pixels of the picture are treated as positive or negative class training samples. Integrative labeling, however, results in heavy mixture of training samples. thus traditional generative density estimation methods can't be used directly in that they perform poorly with heavily polluted training samples. In this paper, by utilizing the prior knowledge of high separability between positive and negative class color distributions, a discriminative learning based GMM(DiscGMM) is proposed for integrative labeling. Besides generating the polluted positive-class samples with comparatively high probability, optimal parameters found by DiscGMM also enjoy a comparatively low probability of generating negative-class samples. the parameter learning problem is solved by a modified Expectation Maximization (EM) algorithm. In an integrative labeling experiment of skin detection, DiscGMM is testified to enjoy much better performance than generative density estimation methods and shows qualified results.
Ubiquitous computing increases the pressure on the software industry to produce ever more and error-free code. Two recipes from automated programming are available to meet this challenge: On the one hand, generative p...
ISBN:
(纸本)9780769521312
Ubiquitous computing increases the pressure on the software industry to produce ever more and error-free code. Two recipes from automated programming are available to meet this challenge: On the one hand, generative programming raises the level of abstraction in software development by describing problems in high-level domain-specific languages and making them executable. On the other hand, in situations where one needs to produce a family of similar programs, product line engineering supports code reuse by composing programs from a set of common assets (or features). AHEAD (Algebraic Hierarchical Equations for Application Design) is a framework for generative programming and product line engineering that achieves additional productivity gains by scaling feature composition up. Our contribution is GRAFT, a calculus that gives a formal foundation to AHEAD and provides several mechanisms for making sure that feature combinations are legal and that features in themselves are consistent.
We present a correlated and gate which may be used to propagate uncertainty and dependence through Boolean functions, since any Boolean function may be expressed as a combination of and and not operations. We argue th...
详细信息
this study is concerned withthe evaluation of wind power projects under the Clean Development Mechanism (CDM), not only for the purpose of CDM verification, but also for the financing of the project. A real options m...
详细信息
this study is concerned withthe evaluation of wind power projects under the Clean Development Mechanism (CDM), not only for the purpose of CDM verification, but also for the financing of the project. A real options model is developed in this paper to evaluate the investment decisions on wind power project. the model obtains the real value of the project and determines the optimal time to invest wind power project. Stochastic programming is employed to evaluate the real options model, and a scenario tree, generated by path-based sampling method and LHS discretisation, is constructed to approximate the original stochastic program.
In this paper, a combination ANN/Fuzzy techniques are used to design a Novel Fuzzy Single Neuron PID (NFSNPID) controller to achieve high performance brushless DC motor. the design steps include two parts. the first p...
详细信息
ISBN:
(纸本)9781538609903
In this paper, a combination ANN/Fuzzy techniques are used to design a Novel Fuzzy Single Neuron PID (NFSNPID) controller to achieve high performance brushless DC motor. the design steps include two parts. the first part uses the genetic algorithm (GA) to obtain the optimum parameters of Single Neuron PID (SNPID) controller, while the former deals withthe design of fuzzy logic control to update the weights of SNPID control online. To demonstrate the designed controller effectiveness, a comparative study is made with between the NFSNPID, Conventional Fuzzy Single Neuron PID CFSNPID and SNPID. All controllers were used to drive, separately, the brushless DC motor against the sudden change of load and operating speed. the performed simulations show better results that motivate for further investigations.
暂无评论