Littérature scientifique sur le sujet « Matrix pseudoinversion »

Many researchers have investigated the time-varying (TV) matrix pseudoinverse problem in recent years, for its importance in addressing TV problems in science and engineering. In this paper, the problem of calculating the inverse or pseudoinverse of an arbitrary TV real matrix is considered and addressed using the singular value decomposition (SVD) and the zeroing neural network (ZNN) approaches. Since SVD is frequently used to compute the inverse or pseudoinverse of a matrix, this research proposes a new ZNN model based on the SVD method as well as the technique of Tikhonov regularization, for solving the problem in continuous time. Numerical experiments, involving the pseudoinversion of square, rectangular, singular, and nonsingular input matrices, indicate that the proposed models are effective for solving the problem of the inversion or pseudoinversion of time varying matrices.

Many researchers have addressed problems involving time-varying (TV) general linear matrix equations (GLMEs) because of their importance in science and engineering. This research discusses and solves the topic of solving TV GLME using the zeroing neural network (ZNN) design. Five new ZNN models based on novel error functions arising from gradient-descent and Newton optimization methods are presented and compared to each other and to the standard ZNN design. Pseudoinversion is involved in four proposed ZNN models, while three of them are related to Newton’s optimization method. Heterogeneous numerical examples show that all models successfully solve TV GLMEs, although their effectiveness varies and depends on the input matrix.

This study analyses the robustness and convergence characteristics of a neural network. First, a special class of recurrent neural network (RNN), termed a continuous-time Zhang neural network (CTZNN) model, is presented and investigated for dynamic matrix pseudoinversion. Theoretical analysis of the CTZNN model demonstrates that it has good robustness against various types of noise. In addition, considering the requirements of digital implementation and online computation, the optimal sampling gap for a discrete-time Zhang neural network (DTZNN) model under noisy environments is proposed. Finally, experimental results are presented, which further substantiate the theoretical analyses and demonstrate the effectiveness of the proposed ZNN models for computing a dynamic matrix pseudoinverse under noisy environments.

The computation of the time-varying matrix pseudoinverse has become crucial in recent years for solving time-varying problems in engineering and science domains. This paper investigates the issue of calculating the time-varying pseudoinverse based on full-rank decomposition (FRD) using the zeroing neural network (ZNN) method, which is currently considered to be a cutting edge method for calculating the time-varying matrix pseudoinverse. As a consequence, for the first time in the literature, a new ZNN model called ZNNFRDP is introduced for time-varying pseudoinversion and it is based on FRD. FourFive numerical experiments investigate and confirm that the ZNNFRDP model performs as well as, if not better than, other well-performing ZNN models in the calculation of the time-varying pseudoinverse. Additionally, theoretical analysis and numerical findings have both supported the effectiveness of the proposed model.

Negli ultimi anni l’utilizzo di dispositivi aptici, atti cioè a riprodurre l’interazione fisica con l’ambiente remoto o virtuale, si sta diffondendo in vari ambiti della robotica e dell’informatica, dai videogiochi alla chirurgia robotizzata eseguita in teleoperazione, dai cellulari alla riabilitazione. In questo lavoro di tesi abbiamo voluto considerare nuovi punti di vista sull’argomento, allo scopo di comprendere meglio come riportare l’essere umano, che è l’unico fruitore del ritorno di forza, tattile e di telepresenza, al centro della ricerca sui dispositivi aptici. Allo scopo ci siamo focalizzati su due aspetti: una manipolazione del segnale di forza mutuata dalla percezione umana e l’utilizzo di architetture multicore per l’implementazione di algoritmi aptici e robotici. Con l’aiuto di un setup sperimentale creato ad hoc e attraverso l’utilizzo di un joystick con ritorno di forza a 6 gradi di libertà, abbiamo progettato degli esperimenti psicofisici atti all’identificazione di soglie differenziali di forze/coppie nel sistema mano-braccio. Sulla base dei risultati ottenuti abbiamo determinato una serie di funzioni di scalatura del segnale di forza, una per ogni grado di libertà, che permettono di aumentare l’abilità umana nel discriminare stimoli differenti. L’utilizzo di tali funzioni, ad esempio in teleoperazione, richiede la possibilità di variare il segnale di feedback e il controllo del dispositivo sia in relazione al lavoro da svolgere, sia alle peculiari capacità dell’utilizzatore. La gestione del dispositivo deve quindi essere in grado di soddisfare due obbiettivi tendenzialmente in contrasto, e cioè il raggiungimento di alte prestazioni in termini di velocità, stabilità e precisione, abbinato alla flessibilità tipica del software. Una soluzione consiste nell’affidare il controllo del dispositivo ai nuovi sistemi multicore che si stanno sempre più prepotentemente affacciando sul panorama informatico. Per far ciò una serie di algoritmi consolidati deve essere portata su sistemi paralleli. In questo lavoro abbiamo dimostrato che è possibile convertire facilmente vecchi algoritmi già implementati in hardware, e quindi intrinsecamente paralleli. Un punto da definire rimane però quanto costa portare degli algoritmi solitamente descritti in VLSI e schemi in un linguaggio di programmazione ad alto livello. Focalizzando la nostra attenzione su un problema specifico, la pseudoinversione di matrici che è presente in molti algoritmi di dinamica e cinematica, abbiamo mostrato che un’attenta progettazione e decomposizione del problema permette una mappatura diretta sulle unità di calcolo disponibili. In aggiunta, l’uso di parallelismo a livello di dati su macchine SIMD permette di ottenere buone prestazioni utilizzando semplici operazioni vettoriali come addizioni e shift. Dato che di solito tali istruzioni fanno parte delle implementazioni hardware la migrazione del codice risulta agevole. Abbiamo testato il nostro approccio su una Sony PlayStation 3 equipaggiata con un processore IBM Cell Broadband Engine.
In the last years the use of haptic feedback has been used in several applications, from mobile phones to rehabilitation, from video games to robotic aided surgery. The haptic devices, that are the interfaces that create the stimulation and reproduce the physical interaction with virtual or remote environments, have been studied, analyzed and developed in many ways. Every innovation in the mechanics, electronics and technical design of the device it is valuable, however it is important to maintain the focus of the haptic interaction on the human being, who is the only user of force feedback. In this thesis we worked on two main topics that are relevant to this aim: a perception based force signal manipulation and the use of modern multicore architectures for the implementation of the haptic controller. With the help of a specific experimental setup and using a 6 dof haptic device we designed a psychophysical experiment aimed at identifying of the force/torque differential thresholds applied to the hand-arm system. On the basis of the results obtained we determined a set of task dependent scaling functions, one for each degree of freedom of the three-dimensional space, that can be used to enhance the human abilities in discriminating different stimuli. The perception based manipulation of the force feedback requires a fast, stable and configurable controller of the haptic interface. Thus a solution is to use new available multicore architectures for the implementation of the controller, but many consolidated algorithms have to be ported to these parallel systems. Focusing on specific problem, i.e. the matrix pseudoinversion, that is part of the robotics dynamic and kinematic computation, we showed that it is possible to migrate code that was already implemented in hardware, and in particular old algorithms that were inherently parallel and thus not competitive on sequential processors. The main question that still lies open is how much effort is required in order to write these algorithms, usually described in VLSI or schematics, in a modern programming language. We show that a careful task decomposition and design permit a mapping of the code on the available cores. In addition, the use of data parallelism on SIMD machines can give good performance when simple vector instructions such as add and shift operations are used. Since these instructions are present also in hardware implementations the migration can be easily performed. We tested our approach on a Sony PlayStation 3 game console equipped with IBM Cell Broadband Engine processor.