This paper explores a pragmatic approach to multiple object tracking where the main focus is associate objects efficiently for online and realtime applications. To this end, detection quality identified as key factor influencing performance, changing detector can improve by up 18.9%. Despite only using rudimentary combination of familiar techniques such Kalman Filter Hungarian algorithm components, achieves an accuracy comparable state-of-the-art trackers. Furthermore, due simplicity our...

Malicious software (malware) has been extensively used for illegal activity and new malware variants are discovered at an alarmingly high rate. The ability to group into families with similar characteristics makes possible create mitigation strategies that work a whole class of programs. In this paper, we present family classification approach using deep neural network based on the ResNet-50 architecture. Malware samples represented as byteplot grayscale images is trained freezing...

We introduce a new statistical modelling technique for building occupancy maps. The problem of mapping is addressed as classification task where the robot’s environment classified into regions and free space. This obtained by employing modified Gaussian process non-parametric Bayesian learning to exploit fact that real-world environments inherently possess structure. structure introduces dependencies between points on map which are not accounted many common techniques such grids. Our...

Environmental Monitoring (EM) is typically performed using sensor networks that collect measurements in predefined static locations. The possibility of having one or more autonomous robots to perform this task increases versatility and reduces the number necessary nodes cover same area. However, several problems arise when making use moving for EM. main challenges are how build an accurate spatial-temporal model while choosing locations measuring phenomenon. This paper addresses problem by...

Abstract Building a model of large‐scale terrain that can adequately handle uncertainty and incompleteness in statistically sound way is challenging problem. This work proposes the use Gaussian processes as models terrain. The proposed naturally provides multiresolution representation space, incorporates handles uncertainties aptly, copes with sensory information. process regression techniques are applied to estimate interpolate (to fill gaps occluded areas) elevation information across...

Environmental monitoring with mobile robots requires solving the informative path planning problem. A key challenge is how to compute a continuous over space and time that will allow robot best sample environment for an initially unknown phenomenon. To address this problem we devise layered Bayesian Optimisation approach uses two Gaussian Processes, one model phenomenon other quality of selected paths. By using different acquisition functions both models tackle exploration-exploitation trade...

The vast amount of data robots can capture today motivates the development fast and scalable statistical tools to model environment robot operates in.We devise a new technique for representation through continuous occupancy mapping that improves on popular grip maps in two fundamental aspects: 1) it does not assume an priori discretisation world into grid cells therefore provide at arbitrary resolution; 2) captures relationships between measurements naturally, thus being more robust outliers...

The vast amount of data robots can capture today motivates the development fast and scalable statistical tools to model space robot operates in. We devise a new technique for environment representation through continuous occupancy mapping that improves on popular grip maps in two fundamental aspects: (1) it does not assume an priori discrimination world into grid cells therefore provide at arbitrary resolution; (2) captures spatial relationships between measurements naturally, thus being...

We introduce BayesSim 1 , a framework for robotics simulations allowing full Bayesian treatment the parameters of simulator.As simulators become more sophisticated and able to represent dynamics accurately, fundamental problems in such as motion planning perception can be solved simulation solutions transferred physical robot.However, even most complex simulator might still not reality all its details either due inaccurate parametrization or simplistic assumptions dynamic models.BayesSim...

This paper investigates the possibility of recognising individual persons from their walking gait using three-dimensional 'skeleton' data an inexpensive consumer-level sensor, Microsoft 'Kinect'. In experimental pilot study it is shown that K-means algorithm - as a candidate unsupervised clustering able to cluster samples four with nett accuracy 43.6%.

Abstract This paper presents a framework for integrating sensor information from an inertial measuring unit (IMU), global positioning system (GPS) receiver, and monocular vision camera mounted to low‐flying unmanned aerial vehicle (UAV) producing large‐scale terrain reconstructions classifying different species of vegetation within the environment. The reconstruction phase integrates all using statistically optimal nonlinear least‐squares bundle adjustment algorithm estimate poses...

Popular computational models of decision-making make specific assumptions about learning processes that may cause them to underfit observed behaviours. Here we suggest an alternative method using recurrent neural networks (RNNs) generate a flexible family have sufficient capacity represent the complex and decision- making strategies used by humans. In this approach, RNN is trained predict next action subject will take in task and, way, learns imitate underlying subjects' choices their...

Traditional robotic approaches rely on an accurate model of the environment, a detailed description how to perform task, and robust perception system keep track current state. On other hand, reinforcement learning can operate directly from raw sensory inputs with only reward signal describe but are extremely sampleinefficient brittle. In this work, we combine strengths model-based methods flexibility learning-based obtain general method that is able overcome inaccuracies in robotics...

Learning from offline task demonstrations is a problem of great interest in robotics. For simple short-horizon manipulation tasks with modest variation instances, learning small set can produce controllers that successfully solve the task. However, leveraging fixed batch data be problematic for larger datasets and longer-horizon greater variations. The exhibit substantial diversity consist suboptimal solution approaches. In this paper, we propose Implicit Reinforcement without Interaction at...

Robotic cutting of soft materials is critical for applications such as food processing, household automation, and surgical manipulation. As in other areas robotics, simulators can facilitate controller verification, policy learning, dataset generation. Moreover, differentiable enable gradient-based optimization, which invaluable calibrating simulation parameters optimizing controllers. In this work, we present DiSECt: the first simulator materials. The augments finite element method (FEM)...

The need for efficient monitoring of spatio-temporal dynamics in large environmental surveillance applications motivates the use robotic sensors to achieve sufficient spatial and temporal coverage. A common approach machine learning model is nonparametric Bayesian framework known as Gaussian Processes (GPs) (c.f., [1]) which are fully specified by a mean covariance function. However, defining suitable functions that able appropriately complex space-time dependencies environment challenging...

Multi-task learning remains a difficult yet important problem in machine learning. In Gaussian processes the main challenge is definition of valid kernels (covariance functions) able to capture relationships between different tasks. This paper presents novel methodology construct multi-task covariance functions (Mercer kernels) for allowing combination with forms. The method based on Fourier analysis and general arbitrary stationary functions. Analytical solutions cross terms popular forms...

Measurement-while-drilling (MWD) data recorded from drill rigs can provide a valuable estimation of the type and strength rocks being drilled. Typical MWD sensors include bit pressure, rotation pull-down rate, head speed. This letter presents an empirical comparison statistical performance, ease implementation, computational efficiency associated with three machine-learning techniques. A recently proposed method, boosting, is compared two well-established methods, neural networks fuzzy...

In this paper we address the loop closure detection problem in simultaneous localization and mapping ( slam ), present a method for solving using pairwise comparison of point clouds both two three dimensions. The are mathematically described features that capture important geometric statistical properties. used as input to machine learning algorithm AdaBoost, which is build non-linear classifier capable detecting from pairs clouds. Vantage dependency process eliminated by only rotation...

Understanding and predicting how influenza propagates is vital to reduce its impact. In this paper we develop a nonparametric model based on Gaussian process (GP) regression capture the complex spatial temporal dependencies present in data. A stochastic variational inference approach was adopted address scalability. Rather than modeling problem as time-series many studies, space-time by combining different kernels. kernel averaging technique which converts spatially-diffused point processes...

Sensor fusion is a major field in autonomous mobile robots navigation research. These methods integrate information obtained from accelerometers, rate gyros and monocular cameras to provide pose orientation of the robot, which are known literature as Visual-Inertial Simultaneous Localization Mapping systems. For outdoor navigation, sensor algorithms may also use magnetometers GPS modules, since indoor environments certain urban areas they suffer measurements corruption presence ferromagnetic...