Direct load control of a heterogeneous cluster of residential demand flexibility sources is a high-dimensional control problem with partial observability. This work proposes a novel approach that uses a convolutional neural network to extract hidden state-time features to mitigate the curse of partial observability. More specific, a convolutional neural network is used as a function approximator to estimate the state-action value function or Q-function in the supervised learning step of fitted Q-iteration. The approach is evaluated in a qualitative simulation, comprising a cluster of thermostatically controlled loads that only share their air temperature, whilst their envelope temperature remains hidden. The simulation results show that the presented approach is able to capture the underlying hidden features and successfully reduce the electricity cost the cluster.

We present a generalized, 3 dimensional version of the Purcell's swimmer which is a planar mechanism locomoting at low Reynlods number regime. We use Cox theory and resistive force theory to come up with the forces acting on the system. We finally come up with a purely kinematic form of the system's equations.

The objective of this dissertation is to shed light on some fundamental impediments in learning control laws in continuous state spaces. In particular, if one wants to build artificial devices capable to learn motor tasks the same way they learn to classify signals and images, one needs to establish control rules that do not necessitate comparisons between quantities of the surrounding space. We propose, in that context, to take inspiration from the "end effector control" principle, as suggested by neuroscience studies, as opposed to the "displacement control" principle used in the classical control theory.

Nonlinear variational methods have become very powerful tools for many image processing tasks. Recently a new line of research has emerged, dealing with nonlinear eigenfunctions induced by convex functionals. This has provided new insights and better theoretical understanding of convex regularization and introduced new processing methods. However, the theory of nonlinear eigenvalue problems is still at its infancy. We present a new flow that can generate nonlinear eigenfunctions of the form $T(u)=λu$, where $T(u)$ is a nonlinear operator and $λ\in \mathbb{R} $ is the eigenvalue. We develop the theory where $T(u)$ is a subgradient element of a regularizing one-homogeneous functional, such as total-variation (TV) or total-generalized-variation (TGV). We introduce two flows: a forward flow and an inverse flow; for which the steady state solution is a nonlinear eigenfunction. The forward flow monotonically smooths the solution (with respect to the regularizer) and simultaneously increases the $L^2$ norm. The inverse flow has the opposite characteristics. For both flows, the steady state depends on the initial condition, thus different initial conditions yield different eigenfunctions. This enables a deeper investigation into the space of nonlinear eigenfunctions, allowing to produce numerically diverse examples, which may be unknown yet. In addition we suggest an indicator to measure the affinity of a function to an eigenfunction and relate it to pseudo-eigenfunctions in the linear case.

A main problem in autonomous vehicles in general, and in \acp{UAV} in particular, is the determination of the attitude angles. A novel method to estimate these angles using off-the-shelf components is presented. This paper introduces an \ac{AHRS} based on the \ac{UKF} using the \ac{TRIAD} algorithm as the observation model. The performance of the method is assessed through simulations and compared to an \ac{AHRS} based on the \ac{EKF}. The paper presents field experiment results using a real fixed-wing \ac{UAV}. The results show good real-time performance with low computational cost in a microcontroller.

An autonomous navigation with proven collision avoidance in unknown and dynamic environments is still a challenge, particularly when there are moving obstacles. A popular approach to collision avoidance in the face of moving obstacles is based on model predictive algorithms, which, however, may be computationally expensive. Hence, we adopt a reactive potential field approach here. At every cycle, the proposed approach requires only current robot states relative to the closest obstacle point to find the potential field in the current position; thus, it is more computationally efficient and more suitable to scale up for multiple agent scenarios. Our main contribution here is to write the reactive potential field based motion controller as a hybrid automaton, and then formally verify its safety using differential dynamic logic. In particular, we can guarantee a passive safety property, which means that collisions cannot occur if the robot is to blame, namely a collision can occur only if the robot is at rest. The proposed controller and verification results are demonstrated via simulations and implementation on a Pioneer P3-AT robot.

Despite the great success of using gradient-based controllers to stabilize rigid formations of autonomous agents in the past years, surprising yet intriguing undesirable collective motions have been reported recently when inconsistent measurements are used in the agents' local controllers. To make the existing gradient control robust against such measurement inconsistency, we exploit local estimators following the well known internal model principle for robust output regulation control. The new estimator-based gradient control is still distributed in nature and can be constructed systematically even when the number of agents in a rigid formation grows. We prove rigorously that the proposed control is able to guarantee exponential convergence and then demonstrate through robotic experiments and computer simulations that the reported inconsistency-induced orbits of collective movements are effectively eliminated.

Recent progress in randomized motion planners has led to the development of a new class of sampling-based algorithms that provide asymptotic optimality guarantees, notably the RRT* and the PRM* algorithms. Careful analysis reveals that the so-called "rewiring" step in these algorithms can be interpreted as a local policy iteration (PI) step (i.e., a local policy evaluation step followed by a local policy improvement step) so that asymptotically, as the number of samples tend to infinity, both algorithms converge to the optimal path almost surely (with probability 1). Policy iteration, along with value iteration (VI) are common methods for solving dynamic programming (DP) problems. Based on this observation, recently, the RRT$^{\#}$ algorithm has been proposed, which performs, during each iteration, Bellman updates (aka "backups") on those vertices of the graph that have the potential of being part of the optimal path (i.e., the "promising" vertices). The RRT$^{\#}$ algorithm thus utilizes dynamic programming ideas and implements them incrementally on randomly generated graphs to obtain high quality solutions. In this work, and based on this key insight, we explore a different class of dynamic programming algorithms for solving shortest-path problems on random graphs generated by iterative sampling methods. These class of algorithms utilize policy iteration instead of value iteration, and thus are better suited for massive parallelization. Contrary to the RRT* algorithm, the policy improvement during the rewiring step is not performed only locally but rather on a set of vertices that are classified as "promising" during the current iteration. This tends to speed-up the whole process. The resulting algorithm, aptly named Policy Iteration-RRT$^{\#}$ (PI-RRT$^{\#}$) is the first of a new class of DP-inspired algorithms for randomized motion planning that utilize PI methods.

Brains and sensory systems evolved to guide motion. Central to this task is controlling the approach to stationary obstacles and detecting moving organisms. Looming has been proposed as the main monocular visual cue for detecting the approach of other animals and avoiding collisions with stationary obstacles. Elegant neural mechanisms for looming detection have been found in the brain of insects and vertebrates. However, looming has not been analyzed in the context of collisions between two moving animals. We propose an alternative strategy, Generalized Regressive Motion (GRM), which is consistent with recently observed behavior in fruit flies. Geometric analysis proves that GRM is a reliable cue to collision among conspecifics, whereas agent-based modeling suggests that GRM is a better cue than looming as a means to detect approach, prevent collisions and maintain mobility.

The SLAM problem is known to have a special property that when robot orientation is known, estimating the history of robot poses and feature locations can be posed as a standard linear least squares problem. In this work, we develop a SLAM framework that uses relative feature-to-feature measurements to exploit this structural property of SLAM. Relative feature measurements are used to pose a linear estimation problem for pose-to-pose orientation constraints. This is followed by solving an iterative non-linear on-manifold optimization problem to compute the maximum likelihood estimate for robot orientation given relative rotation constraints. Once the robot orientation is computed, we solve a linear problem for robot position and map estimation. Our approach reduces the computational burden of non-linear optimization by posing a smaller optimization problem as compared to standard graph-based methods for feature-based SLAM. Further, empirical results show our method avoids catastrophic failures that arise in existing methods due to using odometery as an initial guess for non-linear optimization, while its accuracy degrades gracefully as sensor noise is increased. We demonstrate our method through extensive simulations and comparisons with an existing state-of-the-art solver.

In this paper we address the problem of robot movement adaptation under various environmental constraints interactively. Motion primitives are generally adopted to generate target motion from demonstrations. However, their generalization capability is weak while facing novel environments. Additionally, traditional motion generation methods do not consider the versatile constraints from various users, tasks, and environments. In this work, we propose a co-active learning framework for learning to adapt robot end-effector's movement for manipulation tasks. It is designed to adapt the original imitation trajectories, which are learned from demonstrations, to novel situations with various constraints. The framework also considers user's feedback towards the adapted trajectories, and it learns to adapt movement through human-in-the-loop interactions. The implemented system generalizes trained motion primitives to various situations with different constraints considering user preferences. Experiments on a humanoid platform validate the effectiveness of our approach.

Given a musical audio recording, the goal of automatic music transcription is to determine a score-like representation of the piece underlying the recording. Despite significant interest within the research community, several studies have reported on a 'glass ceiling' effect, an apparent limit on the transcription accuracy that current methods seem incapable of overcoming. In this paper, we explore how much this effect can be mitigated by focusing on a specific instrument class and making use of additional information on the recording conditions available in studio or home recording scenarios. In particular, exploiting the availability of single note recordings for the instrument in use we develop a novel signal model employing variable-length spectro-temporal patterns as its central building blocks - tailored for pitched percussive instruments such as the piano. Temporal dependencies between spectral templates are modeled, resembling characteristics of factorial scaled hidden Markov models (FS-HMM) and other methods combining Non-Negative Matrix Factorization with Markov processes. In contrast to FS-HMMs, our parameter estimation is developed in a global, relaxed form within the extensible alternating direction method of multipliers (ADMM) framework, which enables the systematic combination of basic regularizers propagating sparsity and local stationarity in note activity with more complex regularizers imposing temporal semantics. The proposed method achieves an f-measure of 93-95% for note onsets on pieces recorded on a Yamaha Disklavier (MAPS DB).

In this paper we address the inconsistency of the EKF-based SLAM algorithm that stems from non-observability of the origin and orientation of the global reference frame. We prove on the non-linear two-dimensional problem with point landmarks observed that this type of inconsistency is remedied using the Invariant EKF, a recently introduced variant ot the EKF meant to account for the symmetries of the state space. Extensive Monte-Carlo runs illustrate the theoretical results.

An essential function for automated vehicle technologies is accurate localization. It is difficult, however, to achieve lane-level accuracy with low-cost Global Navigation Satellite System (GNSS) receivers due to the biased noisy pseudo-range measurements. Approaches such as Differential GNSS can improve the accuracy, but usually require an enormous amount of investment in base stations. The emerging connected vehicle technologies provide an alternative approach to improving the localization accuracy. It has been shown in this paper that localization accuracy can be enhanced by fusing GNSS information within a group of connected vehicles and matching the configuration of the group to a digital map to eliminate the common bias in localization. A Rao-Blackwellized particle filter (RBPF) was used to jointly estimate the common biases of the pseudo-ranges and the vehicles positions. Multipath biases, which are non-common to vehicles, were mitigated by a multi-hypothesis detection-rejection approach. The temporal correlation was exploited through the prediction-update process. The proposed approach was compared to the existing static and smoothed static methods in the intersection scenario. Simulation results show that the proposed algorithm reduced the estimation error by fifty percent and reduced the estimation variance by two orders of magnitude.

Anthropomimetic robots are robots that sense, behave, interact and feel like humans. By this definition, anthropomimetic robots require human-like physical hardware and actuation, but also brain-like control and sensing. The most self-evident realization to meet those requirements would be a human-like musculoskeletal robot with a brain-like neural controller. While both musculoskeletal robotic hardware and neural control software have existed for decades, a scalable approach that could be used to build and control an anthropomimetic human-scale robot has not been demonstrated yet. Combining Myorobotics, a framework for musculoskeletal robot development, with SpiNNaker, a neuromorphic computing platform, we present the proof-of-principle of a system that can scale to dozens of neurally-controlled, physically compliant joints. At its core, it implements a closed-loop cerebellar model which provides real-time low-level neural control at minimal power consumption and maximal extensibility: higher-order (e.g., cortical) neural networks and neuromorphic sensors like silicon-retinae or -cochleae can naturally be incorporated.

Developing intelligent systems requires combining results from both industry and academia. In this report you find an overview of relevant research fields and industrially applicable technologies for building very large scale cyber physical systems. A concept architecture is used to illustrate how existing pieces may fit together, and the maturity of the subsystems is estimated. The goal is to structure the developments and the challenge of machine intelligence for Consumer and Industrial Internet technologists, cyber physical systems researchers and people interested in the convergence of data & Internet of Things. It can be used for planning developments of intelligent systems.

In the context of change-point detection, addressed by Total Variation minimization strategies, an efficient on-the-fly algorithm has been designed leading to exact solutions for univariate data. In this contribution, an extension of such an on-the-fly strategy to multivariate data is investigated. The proposed algorithm relies on the local validation of the Karush-Kuhn-Tucker conditions on the dual problem. Showing that the non-local nature of the multivariate setting precludes to obtain an exact on-the-fly solution, we devise an on-the-fly algorithm delivering an approximate solution, whose quality is controlled by a practitioner-tunable parameter, acting as a trade-off between quality and computational cost. Performance assessment shows that high quality solutions are obtained on-the-fly while benefiting of computational costs several orders of magnitude lower than standard iterative procedures. The proposed algorithm thus provides practitioners with an efficient multivariate change-point detection on-the-fly procedure.

Recently it has been reported that range-measurement inconsistency, or equivalently mismatches in prescribed inter-agent distances, may prevent the popular gradient controllers from guiding rigid formations of mobile agents to converge to their desired shape, and even worse from standing still at any location. In this paper, instead of treating mismatches as the source of ill performance, we take them as design parameters and show that by introducing such a pair of parameters per distance constraint, distributed controller achieving simultaneously both formation and motion control can be designed that not only encompasses the popular gradient control, but more importantly allows us to achieve constant collective translation, rotation or their combination while guaranteeing asymptotically no distortion in the formation shape occurs. Such motion control results are then applied to (a) the alignment of formations orientations and (b) enclosing and tracking a moving target. Besides rigorous mathematical proof, experiments using mobile robots are demonstrated to show the satisfying performances of the proposed formation-motion distributed controller.

Network controllers (NCs) are devices that are capable of converting dynamic, spatially extended, and functionally specialized modules into a taskable goal-oriented group called networked control system. This paper examines the practical aspects of designing and building an NC that uses the Internet as a communication medium. It focuses on finding compatible controller components that can be integrated via a host structure in a manner that makes it possible to network, in real-time, a webcam, an unmanned ground vehicle (UGV), and a remote computer server along with the necessary operator software interface. The aim is to deskill the UGV navigation process and yet maintain a robust performance. The structure of the suggested controller, its components, and the manner in which they are interfaced are described. Thorough experimental results along with performance assessment and comparisons to a previously implemented NC are provided.

The main objective of this paper is to provide a tool for performing path planning at the servo level of a mobile robot. The ability to perform, in a provably correct manner, such a complex task at the servo level can lead to a large increase in the speed of operation, low energy consumption and high quality of response. Planning has been traditionally limited to the high level controller of a robot. The guidance velocity signal from this stage is usually converted to a control signal using what is known as an electronic speed controller (ESC). This paper demonstrates the ability of the harmonic potential field (HPF) approach to generate a provably correct, constrained, well behaved trajectory and control signal for a rigid, nonholonomic robot in a stationary, cluttered environment. It is shown that the HPF based, servo level planner can address a large number of challenges facing planning in a realistic situation. The suggested approach migrates the rich and provably correct properties of the solution trajectories from an HPF planner to those of the robot. This is achieved using a synchronizing control signal whose aim is to align the velocity of the robot in its local coordinates, with that of the gradient of the HPF. The link between the two is made possible by representing the robot using what the paper terms separable form. The context-sensitive and goal-oriented control signal used to steer the robot is demonstrated to be well behaved and robust in the presence of actuator noise, saturation and uncertainty in the parameters. The approach is developed, proofs of correctness are provided and the capabilities of the scheme are demonstrated using simulation results.

Connectivity maintenance is an essential task in multi-robot systems and it has received a considerable attention during the last years. A connected system can be broken into two or more subsets simply if a single robot fails. A more robust communication can be achieved if the network connectivity is guaranteed in the case of one-robot failures. The resulting network is called biconnected. In \cite{Zareh2016biconnectivitycheck}, we presented a criterion for biconnectivity check, which basically determines a lower bound on the third-smallest eigenvalue of the Laplacian matrix. In this paper, we introduce a decentralized gradient-based protocol to increase the value of the third-smallest eigenvalue of the Laplacian matrix, when the biconnectivity check fails. We also introduce a decentralized algorithm to estimate the eigenvectors of the Laplacian matrix, which are used for defining the gradient. Simulations show the effectiveness of the theoretical findings.

The network connectivity in a group of cooperative robots can be easily broken if one of them loses its connectivity with the rest of the group. In case of having robustness with respect to one-robot-fail, the communication network is termed biconnected. In simple words, to have a biconnected network graph, we need to prove that there exists no articulation point. We propose a decentralized approach that provides sufficient conditions for biconnectivity of the network, and we prove that these conditions are related to the third smallest eigenvalue of the Laplacian matrix. Data exchange among the robots is supposed to be neighbor-to-neighbor.

We introduce a new method for location recovery from pair-wise directions that leverages an efficient convex program that comes with exact recovery guarantees, even in the presence of adversarial outliers. When pairwise directions represent scaled relative positions between pairs of views (estimated for instance with epipolar geometry) our method can be used for location recovery, that is the determination of relative pose up to a single unknown scale. For this task, our method yields performance comparable to the state-of-the-art with an order of magnitude speed-up. Our proposed numerical framework is flexible in that it accommodates other approaches to location recovery and can be used to speed up other methods. These properties are demonstrated by extensively testing against state-of-the-art methods for location recovery on 13 large, irregular collections of images of real scenes in addition to simulated data with ground truth.

Non-negative matrix factorization (NMF) is a common method for generating topic models from text data. NMF is widely accepted for producing good results despite its relative simplicity of implementation and ease of computation. One challenge with applying NMF to large datasets is that intermediate matrix products often become dense, stressing the memory and compute elements of a system. In this article, we investigate a simple but powerful modification of a common NMF algorithm that enforces the generation of sparse intermediate and output matrices. This method enables the application of NMF to large datasets through improved memory and compute performance. Further, we demonstrate empirically that this method of enforcing sparsity in the NMF either preserves or improves both the accuracy of the resulting topic model and the convergence rate of the underlying algorithm.

Nonnegative Matrix Factorization(NMF) is a common used technique in machine learning to extract features out of data such as text documents and images thanks to its natural clustering properties. In particular, it is popular in image processing since it can decompose several pictures and recognize common parts if they're located in the same position over the photos. This paper's aim is to present a way to add the translation invariance to the classical NMF, that is, the algorithms presented are able to detect common features, even when they're shifted, in different original images.

Learning of low-rank matrices is fundamental to many machine learning applications. A state-of-the-art algorithm is the rank-one matrix pursuit (R1MP). However, it can only be used in matrix completion problems with the square loss. In this paper, we develop a more flexible greedy algorithm for generalized low-rank models whose optimization objective can be smooth or nonsmooth, general convex or strongly convex. The proposed algorithm has low per-iteration time complexity and fast convergence rate. Experimental results show that it is much faster than the state-of-the-art, with comparable or even better prediction performance.

We consider a problem of robotic router placement and mobility control with the objective of formation and maintenance of an optimal communication network between a set of transmitter-receiver pairs. In this scenario, the communication path between any transmitter-receiver pair contains a predetermined set of mobile robotic routers nodes. The goal of this work is to design an algorithm to optimize the positions of the robotic nodes to improve the overall performance of the network. We define the optimization metric to be the minimum of the Signal to Interference plus Noise Ratios (SINR) over all the links. In this manuscript, we propose two optimization algorithms to solve this problem in a centralized and a decentralized manner, respectively.We also demonstrate the performances of both algorithms based on a set of simulation experiments.

Autonomous Unmanned Aerial Vehicles (UAVs) have gained popularity due to their many potential application fields. Alongside sophisticated sensors, UAVs can be equipped with communication adaptors aimed for inter-UAV communication. Inter-communication of UAVs to form a UAV swarm raises questions on how to manage its communication structure and mobility. In this paper, we consider therefore the problem of establishing an efficient swarm movement model and a network topology between a collection of UAVs, which are specifically deployed for the scenario of high-quality forest-mapping. The forest environment with its highly heterogeneous distribution of trees and obstacles represents an extreme challenge for a UAV swarm. It requires the swarm to constantly avoid possible collisions with trees, to change autonomously the trajectory, which can lead to disconnection to the swarm, and to reconnect to the swarm after passing the obstacle, while continue collecting environmental data that needs to be fused and assessed efficiently. In this paper, we propose a novel solution to the formation flight problem for UAV swarms. The proposed method provides an adaptive and reliable network structure, which maintains swarm connectivity and communicability. These characteristics are needed to achieve a detailed and accurate description of the environment from the data acquired by the UAV swarm. The main characteristics of our approach are high scalability regarding the number of UAVs in the swarm and the adaptive network topology within the swarm.

This paper extends the capabilities of the harmonic potential field (HPF) approach to planning. The extension covers the situation where the workspace of a robot cannot be segmented into geometrical subregions where each region has an attribute of its own. The suggested approach uses a task-centered, probabilistic descriptor of the workspace as an input to the planner. This descriptor is processed, along with a goal point, to yield the navigation policy needed to steer the agent from any point in its workspace to the target. The approach is easily adaptable to planning in a cluttered environment containing a vector drift field. The extension of the HPF approach is based on the physical analogy with an electric current flowing in a nonhomogeneous conducting medium. The resulting potential field is known as the gamma-harmonic potential (GHPF). Proofs of the ability of the modified approach to avoid zero-probability (definite threat) regions and to converge to the goal are provided. The capabilities of the planer are demonstrated using simulation.

Spatial multi-agency has been receiving growing attention from researchers exploring many of the aspects and modalities of this phenomenon. The aim is to develop the theoretical background needed for a multitude of applications involving the sharing of resources by more than one agent. A traffic management system is one of these applications. Here, a large group of mobile robots that are operating in communication-limited, and sensory-limited modes are required to cope with each others presence as well as the contents of their environment while preserving their ability to reach their preset, independent goals. This work explores the construction of a decentralized traffic controller for a large group of agents sharing a workspace with stationary forbidden regions. The suggested multi-agent motion controller is complete provided that a lenient condition on the geometry of the workspace is upheld. It has a low computational effort that linearly increases with the number of agents. The controller is also self-organizing; therefore, it is able to deal, on its own, with incomplete information and unexpected situations. In addition to the above, the controller has an open structure to enable any agent to join or leave the group without the remaining agents having to adjust the manner in which they function. To meet these requirements, a definition of decentralization is suggested. This definition equates decentralization to self-organization in a group of agents operating in an artificial life mode. The definition is used to provide guidelines for the construction of the multi-agent controller. The controller is realized using the potential field approach. Theoretical developments, as well as simulation results, are provided.