This work introduces the concept of tangent space regularization for neural-network models of dynamical systems. The tangent space to the dynamics function of many physical systems of interest in control applications exhibits useful properties, e.g., smoothness, motivating regularization of the model Jacobian along system trajectories using assumptions on the tangent space of the dynamics. Without assumptions, large amounts of training data are required for a neural network to learn the full non-linear dynamics without overfitting. We compare different network architectures on one-step prediction and simulation performance and investigate the propensity of different architectures to learn models with correct input-output Jacobian. Furthermore, the influence of $L_2$ weight regularization on the learned Jacobian eigenvalue spectrum, and hence system stability, is investigated.

Active inference is an ambitious theory that treats perception, inference and action selection of autonomous agents under the heading of a single principle. It suggests biologically plausible explanations for many cognitive phenomena, including consciousness. In active inference, action selection is driven by an objective function that evaluates possible future actions with respect to current, inferred beliefs about the world. Active inference at its core is independent from extrinsic rewards, resulting in a high level of robustness across e.g.\ different environments or agent morphologies. In the literature, paradigms that share this independence have been summarised under the notion of intrinsic motivations. In general and in contrast to active inference, these models of motivation come without a commitment to particular inference and action selection mechanisms. In this article, we study if the inference and action selection machinery of active inference can also be used by alternatives to the originally included intrinsic motivation. The perception-action loop explicitly relates inference and action selection to the environment and agent memory, and is consequently used as foundation for our analysis. We reconstruct the active inference approach, locate the original formulation within, and show how alternative intrinsic motivations can be used while keeping many of the original features intact. Furthermore, we illustrate the connection to universal reinforcement learning by means of our formalism. Active inference research may profit from comparisons of the dynamics induced by alternative intrinsic motivations. Research on intrinsic motivations may profit from an additional way to implement intrinsically motivated agents that also share the biological plausibility of active inference.

We present a method for scalable and fully 3D magnetic field simultaneous localisation and mapping (SLAM) using local anomalies in the magnetic field as a source of position information. These anomalies are due to the presence of ferromagnetic material in the structure of buildings and in objects such as furniture. We represent the magnetic field map using a Gaussian process model and take well-known physical properties of the magnetic field into account. We build local maps using three-dimensional hexagonal block tiling. To make our approach computationally tractable we use reduced-rank Gaussian process regression in combination with a Rao-Blackwellised particle filter. We show that it is possible to obtain accurate position and orientation estimates using measurements from a smartphone, and that our approach provides a scalable magnetic field SLAM algorithm in terms of both computational complexity and map storage.

In recent years, MEMS inertial sensors (3D accelerometers and 3D gyroscopes) have become widely available due to their small size and low cost. Inertial sensor measurements are obtained at high sampling rates and can be integrated to obtain position and orientation information. These estimates are accurate on a short time scale, but suffer from integration drift over longer time scales. To overcome this issue, inertial sensors are typically combined with additional sensors and models. In this tutorial we focus on the signal processing aspects of position and orientation estimation using inertial sensors. We discuss different modeling choices and a selected number of important algorithms. The algorithms include optimization-based smoothing and filtering as well as computationally cheaper extended Kalman filter and complementary filter implementations. The quality of their estimates is illustrated using both experimental and simulated data.

In this paper we provide a finite-sample and an infinite-sample representer theorem for the concatenation of (linear combinations of) kernel functions of reproducing kernel Hilbert spaces. These results serve as mathematical foundation for the analysis of machine learning algorithms based on compositions of functions. As a direct consequence in the finite-sample case, the corresponding infinite-dimensional minimization problems can be recast into (nonlinear) finite-dimensional minimization problems, which can be tackled with nonlinear optimization algorithms. Moreover, we show how concatenated machine learning problems can be reformulated as neural networks and how our representer theorem applies to a broad class of state-of-the-art deep learning methods.

This article presents AutoRally, a 1$:$5 scale robotics testbed for autonomous vehicle research. AutoRally is designed for robustness, ease of use, and reproducibility, so that a team of two people with limited knowledge of mechanical engineering, electrical engineering, and computer science can construct and then operate the testbed to collect real world autonomous driving data in whatever domain they wish to study. Complete documentation to construct and operate the platform is available online along with tutorials, example controllers, and a driving dataset collected at the Georgia Tech Autonomous Racing Facility. Offline estimation algorithms are used to determine parameters for physics-based dynamics models using an adaptive limited memory joint state unscented Kalman filter. Online vehicle state estimation using a factor graph optimization scheme and a convolutional neural network for semantic segmentation of drivable surface are presented. All algorithms are tested with real world data from the fleet of six AutoRally robots at the Georgia Tech Autonomous Racing Facility tracks, and serve as a demonstration of the robot$'$s capabilities.

In recent years, deep reinforcement learning has been shown to be adept at solving sequential decision processes with high-dimensional state spaces such as in the Atari games. Many reinforcement learning problems, however, involve high-dimensional discrete action spaces as well as high-dimensional state spaces. This paper considers entropy bonus, which is used to encourage exploration in policy gradient. In the case of high-dimensional action spaces, calculating the entropy and its gradient requires enumerating all the actions in the action space and running forward and backpropagation for each action, which may be computationally infeasible. We develop several novel unbiased estimators for the entropy bonus and its gradient. We apply these estimators to several models for the parameterized policies, including Independent Sampling, CommNet, Autoregressive with Modified MDP, and Autoregressive with LSTM. Finally, we test our algorithms on two environments: a multi-hunter multi-rabbit grid game and a multi-agent multi-arm bandit problem. The results show that our entropy estimators substantially improve performance with marginal additional computational cost.

Various approaches have been proposed to learn visuo-motor policies for real-world robotic applications. One solution is first learning in simulation then transferring to the real world. In the transfer, most existing approaches need real-world images with labels. However, the labelling process is often expensive or even impractical in many robotic applications. In this paper, we propose an adversarial discriminative sim-to-real transfer approach to reduce the cost of labelling real data. The effectiveness of the approach is demonstrated with modular networks in a table-top object reaching task where a 7 DoF arm is controlled in velocity mode to reach a blue cuboid in clutter through visual observations. The adversarial transfer approach reduced the labelled real data requirement by 50%. Policies can be transferred to real environments with only 93 labelled and 186 unlabelled real images. The transferred visuo-motor policies are robust to novel (not seen in training) objects in clutter and even a moving target, achieving a 97.8% success rate and 1.8 cm control accuracy.

We propose a novel method to accelerate Lloyd's algorithm for K-Means clustering. Unlike previous acceleration approaches that reduce computational cost per iterations or improve initialization, our approach is focused on reducing the number of iterations required for convergence. This is achieved by treating the assignment step and the update step of Lloyd's algorithm as a fixed-point iteration, and applying Anderson acceleration, a well-established technique for accelerating fixed-point solvers. Classical Anderson acceleration utilizes m previous iterates to find an accelerated iterate, and its performance on K-Means clustering can be sensitive to choice of m and the distribution of samples. We propose a new strategy to dynamically adjust the value of m, which achieves robust and consistent speedups across different problem instances. Our method complements existing acceleration techniques, and can be combined with them to achieve state-of-the-art performance. We perform extensive experiments to evaluate the performance of the proposed method, where it outperforms other algorithms in 106 out of 120 test cases, and the mean decrease ratio of computational time is more than 33%.

We introduce a novel approach to Maximum A Posteriori inference based on discrete graphical models. By utilizing local Wasserstein distances for coupling assignment measures across edges of the underlying graph, a given discrete objective function is smoothly approximated and restricted to the assignment manifold. A corresponding multiplicative update scheme combines in a single process (i) geometric integration of the resulting Riemannian gradient flow and (ii) rounding to integral solutions that represent valid labelings. Throughout this process, local marginalization constraints known from the established LP relaxation are satisfied, whereas the smooth geometric setting results in rapidly converging iterations that can be carried out in parallel for every edge.

Small uninhabited aerial vehicles (sUAVs) commonly rely on active propulsion to stay airborne, which limits flight time and range. To address this, autonomous soaring seeks to utilize free atmospheric energy in the form of updrafts (thermals). However, their irregular nature at low altitudes makes them hard to exploit for existing methods. We model autonomous thermalling as a POMDP and present a receding-horizon controller based on it. We implement it as part of ArduPlane, a popular open-source autopilot, and compare it to an existing alternative in a series of live flight tests involving two sUAVs thermalling simultaneously, with our POMDP-based controller showing a significant advantage.

We propose practical algorithms for entrywise $\ell_p$-norm low-rank approximation, for $p = 1$ or $p = \infty$. The proposed framework, which is non-convex and gradient-based, is easy to implement and typically attains better approximations, faster, than state of the art. From a theoretical standpoint, we show that the proposed scheme can attain $(1 + \varepsilon)$-OPT approximations. Our algorithms are not hyperparameter-free: they achieve the desiderata only assuming algorithm's hyperparameters are known a priori---or are at least approximable. I.e., our theory indicates what problem quantities need to be known, in order to get a good solution within polynomial time, and does not contradict to recent inapproximabilty results, as in [46].

We propose and numerically solve a new variational model for automatic saliency detection in digital images. Using a non-local framework we consider a family of edge preserving functions combined with a new quadratic saliency detection term. Such term defines a constrained bilateral obstacle problem for image classification driven by p-Laplacian operators, including the so-called hyper-Laplacian case (0 < p < 1). The related non-convex non-local reactive flows are then considered and applied for glioblastoma segmentation in magnetic resonance fluid-attenuated inversion recovery (MRI-Flair) images. A fast convolutional kernel based approximated solution is computed. The numerical experiments show how the non-convexity related to the hyperLaplacian operators provides monotonically better results in terms of the standard metrics.

We propose a fast second-order method that can be used as a drop-in replacement for current deep learning solvers. Compared to stochastic gradient descent (SGD), it only requires two additional forward-mode automatic differentiation operations per iteration, which has a computational cost comparable to two standard forward passes and is easy to implement. Our method addresses long-standing issues with current second-order solvers, which invert an approximate Hessian matrix every iteration exactly or by conjugate-gradient methods, a procedure that is both costly and sensitive to noise. Instead, we propose to keep a single estimate of the gradient projected by the inverse Hessian matrix, and update it once per iteration. This estimate has the same size and is similar to the momentum variable that is commonly used in SGD. No estimate of the Hessian is maintained. We first validate our method, called CurveBall, on small problems with known closed-form solutions (noisy Rosenbrock function and degenerate 2-layer linear networks), where current deep learning solvers seem to struggle. We then train several large models on CIFAR and ImageNet, including ResNet and VGG-f networks, where we demonstrate faster convergence with no hyperparameter tuning. Code is available.

We describe a new framework for fitting jump models to a sequence of data. The key idea is to alternate between minimizing a loss function to fit multiple model parameters, and minimizing a discrete loss function to determine which set of model parameters is active at each data point. The framework is quite general and encompasses popular classes of models, such as hidden Markov models and piecewise affine models. The shape of the chosen loss functions to minimize determine the shape of the resulting jump model.

Ability for accurate hospital case cost modelling and prediction is critical for efficient health care financial management and budgetary planning. A variety of regression machine learning algorithms are known to be effective for health care cost predictions. The purpose of this experiment was to build an Azure Machine Learning Studio tool for rapid assessment of multiple types of regression models. The tool offers environment for comparing 14 types of regression models in a unified experiment: linear regression, Bayesian linear regression, decision forest regression, boosted decision tree regression, neural network regression, Poisson regression, Gaussian processes for regression, gradient boosted machine, nonlinear least squares regression, projection pursuit regression, random forest regression, robust regression, robust regression with mm-type estimators, support vector regression. The tool presents assessment results arranged by model accuracy in a single table using five performance metrics. Evaluation of regression machine learning models for performing hospital case cost prediction demonstrated advantage of robust regression model, boosted decision tree regression and decision forest regression. The operational tool has been published to the web and openly available for experiments and extensions.

Simultaneous localization and Planning (SLAP) is a crucial ability for an autonomous robot operating under uncertainty. In its most general form, SLAP induces a continuous POMDP (partially-observable Markov decision process), which needs to be repeatedly solved online. This paper addresses this problem and proposes a dynamic replanning scheme in belief space. The underlying POMDP, which is continuous in state, action, and observation space, is approximated offline via sampling-based methods, but operates in a replanning loop online to admit local improvements to the coarse offline policy. This construct enables the proposed method to combat changing environments and large localization errors, even when the change alters the homotopy class of the optimal trajectory. It further outperforms the state-of-the-art FIRM (Feedback-based Information RoadMap) method by eliminating unnecessary stabilization steps. Applying belief space planning to physical systems brings with it a plethora of challenges. A key focus of this paper is to implement the proposed planner on a physical robot and show the SLAP solution performance under uncertainty, in changing environments and in the presence of large disturbances, such as a kidnapped robot situation.

Can a robot grasp an unknown object without seeing it? In this paper, we present a tactile-sensing based approach to this challenging problem of grasping novel objects without prior knowledge of their location or physical properties. Our key idea is to combine touch based object localization with tactile based re-grasping. To train our learning models, we created a large-scale grasping dataset, including more than 30 RGB frames and over 2.8 million tactile samples from 7800 grasp interactions of 52 objects. To learn a representation of tactile signals, we propose an unsupervised auto-encoding scheme, which shows a significant improvement of 4-9% over prior methods on a variety of tactile perception tasks. Our system consists of two steps. First, our touch localization model sequentially 'touch-scans' the workspace and uses a particle filter to aggregate beliefs from multiple hits of the target. It outputs an estimate of the object's location, from which an initial grasp is established. Next, our re-grasping model learns to progressively improve grasps with tactile feedback based on the learned features. This network learns to estimate grasp stability and predict adjustment for the next grasp. Re-grasping thus is performed iteratively until our model identifies a stable grasp. Finally, we demonstrate extensive experimental results on grasping a large set of novel objects using tactile sensing alone. Furthermore, when applied on top of a vision-based policy, our re-grasping model significantly boosts the overall accuracy by 10.6%. We believe this is the first attempt at learning to grasp with only tactile sensing and without any prior object knowledge.

We propose a novel Human-Swarm Interaction (HSI) framework which enables the user to control a swarm shape and formation. The user commands the swarm utilizing just arm gestures and motions which are recorded by an off-the-shelf wearable armband. We propose a novel interpreter system, which acts as an intermediary between the user and the swarm to simplify the user's role in the interaction. The interpreter takes in a high level input drawn using gestures by the user, and translates it into low level swarm control commands. This interpreter employs machine learning, Kalman filtering and optimal control techniques to translate the user input into swarm control parameters. A notion of Human Interpretable dynamics is introduced, which is used by the interpreter for planning as well as to provide feedback to the user. The dynamics of the swarm are controlled using a novel decentralized formation controller based on distributed linear iterations and dynamic average consensus. The framework is demonstrated theoretically as well as experimentally in a 2D environment, with a human controlling a swarm of simulated robots in real time.

We consider Markov Decision Problems defined over continuous state and action spaces, where an autonomous agent seeks to learn a map from its states to actions so as to maximize its long-term discounted accumulation of rewards. We address this problem by considering Bellman's optimality equation defined over action-value functions, which we reformulate into a nested non-convex stochastic optimization problem defined over a Reproducing Kernel Hilbert Space (RKHS). We develop a functional generalization of stochastic quasi-gradient method to solve it, which, owing to the structure of the RKHS, admits a parameterization in terms of scalar weights and past state-action pairs which grows proportionately with the algorithm iteration index. To ameliorate this complexity explosion, we apply Kernel Orthogonal Matching Pursuit to the sequence of kernel weights and dictionaries, which yields a controllable error in the descent direction of the underlying optimization method. We prove that the resulting algorithm, called KQ-Learning, converges with probability 1 to a stationary point of this problem, yielding a fixed point of the Bellman optimality operator under the hypothesis that it belongs to the RKHS. Under constant learning rates, we further obtain convergence to a small Bellman error that depends on the chosen learning rates. Numerical evaluation on the Continuous Mountain Car and Inverted Pendulum tasks yields convergent parsimonious learned action-value functions, policies that are competitive with the state of the art, and exhibit reliable, reproducible learning behavior.

There has been growing interest in extending traditional vector-based machine learning techniques to their tensor forms. An example is the support tensor machine (STM) that utilizes a rank-one tensor to capture the data structure, thereby alleviating the overfitting and curse of dimensionality problems in the conventional support vector machine (SVM). However, the expressive power of a rank-one tensor is restrictive for many real-world data. To overcome this limitation, we introduce a support tensor train machine (STTM) by replacing the rank-one tensor in an STM with a tensor train. Experiments validate and confirm the superiority of an STTM over the SVM and STM.

This paper aims to identify in a practical manner unknown physical parameters, such as mechanical models of actuated robot links, which are critical in dynamical robotic tasks. Key features include the use of an off-the-shelf physics engine and the Bayesian optimization framework. The task being considered is locomotion with a high-dimensional, compliant Tensegrity robot. A key insight, in this case, is the need to project the model identification challenge into an appropriate lower dimensional space for efficiency. Comparisons with alternatives indicate that the proposed method can identify the parameters more accurately within the given time budget, which also results in more precise locomotion control.

This paper presents the CAT Vehicle (Cognitive and Autonomous Test Vehicle) Testbed: a research testbed comprised of a distributed simulation-based autonomous vehicle, with straightforward transition to hardware in the loop testing and execution, to support research in autonomous driving technology. The evolution of autonomous driving technology from active safety features and advanced driving assistance systems to full sensor-guided autonomous driving requires testing of every possible scenario. However, researchers who want to demonstrate new results on a physical platform face difficult challenges, if they do not have access to a robotic platform in their own labs. Thus, there is a need for a research testbed where simulation-based results can be rapidly validated through hardware in the loop simulation, in order to test the software on board the physical platform. The CAT Vehicle Testbed offers such a testbed that can mimic dynamics of a real vehicle in simulation and then seamlessly transition to reproduction of use cases with hardware. The simulator utilizes the Robot Operating System (ROS) with a physics-based vehicle model, including simulated sensors and actuators with configurable parameters. The testbed allows multi-vehicle simulation to support vehicle to vehicle interaction. Our testbed also facilitates logging and capturing of the data in the real time that can be played back to examine particular scenarios or use cases, and for regression testing. As part of the demonstration of feasibility, we present a brief description of the CAT Vehicle Challenge, in which student researchers from all over the globe were able to reproduce their simulation results with fewer than 2 days of interfacing with the physical platform.

Decentralized (PO)MDPs provide an expressive framework for sequential decision making in a multiagent system. Given their computational complexity, recent research has focused on tractable yet practical subclasses of Dec-POMDPs. We address such a subclass called CDEC-POMDP where the collective behavior of a population of agents affects the joint-reward and environment dynamics. Our main contribution is an actor-critic (AC) reinforcement learning method for optimizing CDEC-POMDP policies. Vanilla AC has slow convergence for larger problems. To address this, we show how a particular decomposition of the approximate action-value function over agents leads to effective updates, and also derive a new way to train the critic based on local reward signals. Comparisons on a synthetic benchmark and a real-world taxi fleet optimization problem show that our new AC approach provides better quality solutions than previous best approaches.

Deep learning techniques have been widely applied, achieving state-of-the-art results in various fields of study. This survey focuses on deep learning solutions that target learning control policies for robotics applications. We carry out our discussions on the two main paradigms for learning control with deep networks: deep reinforcement learning and imitation learning. For deep reinforcement learning (DRL), we begin from traditional reinforcement learning algorithms, showing how they are extended to the deep context and effective mechanisms that could be added on top of the DRL algorithms. We then introduce representative works that utilize DRL to solve navigation and manipulation tasks in robotics. We continue our discussion on methods addressing the challenge of the reality gap for transferring DRL policies trained in simulation to real-world scenarios, and summarize robotics simulation platforms for conducting DRL research. For imitation leaning, we go through its three main categories, behavior cloning, inverse reinforcement learning and generative adversarial imitation learning, by introducing their formulations and their corresponding robotics applications. Finally, we discuss the open challenges and research frontiers.

This paper focuses on a similarity measure, known as the Wasserstein distance, with which to compare images. The Wasserstein distance results from a partial differential equation (PDE) formulation of Monge's optimal transport problem. We present an efficient numerical solution method for solving Monge's problem. To demonstrate the measure's discriminatory power when comparing images, we use a $1$-Nearest Neighbour ($1$-NN) machine learning algorithm to illustrate the measure's potential benefits over other more traditional distance metrics and also the Tangent Space distance, designed to perform excellently on the well-known MNIST dataset. To our knowledge, the PDE formulation of the Wasserstein metric has not been presented for dealing with image comparison, nor has the Wasserstein distance been used within the $1$-nearest neighbour architecture.

The search for interpretable reinforcement learning policies is of high academic and industrial interest. Especially for industrial systems, domain experts are more likely to deploy autonomously learned controllers if they are understandable and convenient to evaluate. Basic algebraic equations are supposed to meet these requirements, as long as they are restricted to an adequate complexity. Here we introduce the genetic programming for reinforcement learning (GPRL) approach based on model-based batch reinforcement learning and genetic programming, which autonomously learns policy equations from pre-existing default state-action trajectory samples. GPRL is compared to a straight-forward method which utilizes genetic programming for symbolic regression, yielding policies imitating an existing well-performing, but non-interpretable policy. Experiments on three reinforcement learning benchmarks, i.e., mountain car, cart-pole balancing, and industrial benchmark, demonstrate the superiority of our GPRL approach compared to the symbolic regression method. GPRL is capable of producing well-performing interpretable reinforcement learning policies from pre-existing default trajectory data.

We present a generic framework for spatio-temporal (ST) data modeling, analysis, and forecasting, with a special focus on data that is sparse in both space and time. Our multi-scaled framework is a seamless coupling of two major components: a self-exciting point process that models the macroscale statistical behaviors of the ST data and a graph structured recurrent neural network (GSRNN) to discover the microscale patterns of the ST data on the inferred graph. This novel deep neural network (DNN) incorporates the real time interactions of the graph nodes to enable more accurate real time forecasting. The effectiveness of our method is demonstrated on both crime and traffic forecasting.

Reinforcement learning has emerged as a promising methodology for training robot controllers. However, most results have been limited to simulation due to the need for a large number of samples and the lack of automated-yet-safe data collection methods. Model-based reinforcement learning methods provide an avenue to circumvent these challenges, but the traditional concern has been the mismatch between the simulator and the real world. Here, we show that control policies learned in simulation can successfully transfer to a physical system, composed of three Phantom robots pushing an object to various desired target positions. We use a modified form of the natural policy gradient algorithm for learning, applied to a carefully identified simulation model. The resulting policies, trained entirely in simulation, work well on the physical system without additional training. In addition, we show that training with an ensemble of models makes the learned policies more robust to modeling errors, thus compensating for difficulties in system identification.

Recurrent Neural Networks (RNNs) are becoming increasingly important for time series-related applications which require efficient and real-time implementations. The recent pruning based work ESE suffers from degradation of performance/energy efficiency due to the irregular network structure after pruning. We propose block-circulant matrices for weight matrix representation in RNNs, thereby achieving simultaneous model compression and acceleration. We aim to implement RNNs in FPGA with highest performance and energy efficiency, with certain accuracy requirement (negligible accuracy degradation). Experimental results on actual FPGA deployments shows that the proposed framework achieves a maximum energy efficiency improvement of 35.7$\times$ compared with ESE.