Convolutional Neural Networks (CNNs) are a popular choice for estimating Direction of Arrival (DoA) without explicitly estimating delays between multiple microphones. The CNN method first optimises unknown filter weights (of a CNN) by using observations and ground-truth directional information. This trained CNN is then used to predict incident directions given test observations. Most existing methods train using spectrally-flat random signals and test using speech. In this paper, which focuses on single source DoA estimation, we find that training with speech or music signals produces a relative improvement in DoA accuracy for a variety of audio classes across 16 acoustic conditions and 9 DoAs, amounting to an average improvement of around 17% and 19% respectively when compared to training with spectrally flat random signals. This improvement is also observed in scenarios in which the speech and music signals are synthesised using, for example, a Generative Adversarial Network (GAN). When the acoustic environments during test and training are similar and reverberant, training a CNN with speech outperforms Generalized Cross Correlation (GCC) methods by about 125%. When the test conditions are different, a CNN performs comparably. This paper takes a step towards answering open questions in the literature regarding the nature of the signals used during training, as well as the amount of data required for estimating DoA using CNNs.

Acoustic source localization (ASL) is an important problem. Despite much attention over the past few decades, rapid and robust ASL still remains elusive. A popular approach is to use a circular array of microphones to record the acoustic signal followed by some form of optimization to deduce the most likely location of the source. In this paper, we study the impact of the configuration of microphones on the accuracy of localization. We perform experiments using simulation as well as real measurements using a 72-microphone acoustic camera which confirm that circular configurations lead to higher localization error than spiral and wheel configurations when considering large regions of space. Moreover, the configuration of choice is intricately tied to the optimization scheme. We show that direct optimization of well known formulations for ASL yield errors similar to the state of the art (steered response power) with 6 × less computation.

This paper proposes a semantic world model framework for hierarchical distributed representation of knowledge in autonomous underwater systems. This framework aims to provide a more capable and holistic system, involving semantic interoperability among all involved information sources. This will enhance interoperability, independence of operation, and situation awareness of the embedded service-oriented agents for autonomous platforms. The results obtained specifically affect the mission flexibility, robustness, and autonomy. The presented framework makes use of the idea that heterogeneous real-world data of very different type must be processed by (and run through) several different layers, to be finally available in a suited format and at the right place to be accessible by high-level decision-making agents. In this sense, the presented approach shows how to abstract away from the raw real-world data step by step by means of semantic technologies. The paper concludes by demonstrating the benefits of the framework in a real scenario. A hardware fault is simulated in a REMUS 100 AUV while performing a mission. This triggers a knowledge exchange between the status monitoring agent and the adaptive mission planner embedded agent. By using the proposed framework, both services can interchange information while remaining domain independent during their interaction with the platform. The results of this paper are readily applicable to land and air robotics.

The multi-layer perceptron (MLP) classifier has excellent discriminatory properties but forms open decision boundaries, which makes it inappropriate for detecting nonclass data. The auto-association neural network (AANN), on the other hand, creates closed decision boundaries around the training set and is thus appropriate for detection and verification in the absence of counter-examples. However, we illustrate that AANNs may fall short in discriminating between classes that lie close to each other or are overlapping in feature space. To overcome each of the network types' weaknesses, we propose a combined system consisting of one MLP and C AANNs for C-class recognition problems. Experimental results show that we can maintain good discriminatory properties whilst reliably detecting non-class data. This is illustrated in the context of radio communication signal recognition.

Hybrid systems using evolution to optimize neural network design or training are usually limited in scope and effectiveness. A system is presented that permits the widest variety of networks to be evolved using a two-stage GA approach. Networks generated for a benchmark machine learning task compare favourably with alternative methods.

The need for embedding fault diagnosis into goal-orientated autonomous robotic vehicles for increased mission robustness is described. The RECOVERY system, a method for increasing the diagnostic capability by integrating commonly available heterogeneous knowledge is presented. Initial real-water results using the Ocean Systems Laboratory's RAUVER vehicle are given.

The application of two distinct approaches to the diagnosis of faults in digital communication equipment is examined. The selected artificial-intelligence approaches use either rule-based or machine-learning techniques. Faults up to 8-dB TWT (traveling-wave tube) overdrive, 10% spacing error in the signal constellation, and 5 degrees nonorthogonality in the modulating carriers are introduced on an 11-GHz radio. Each approach shown is capable of diagnosing both the type and magnitude of the introduced faults, subject to certain constraints for each system. Results indicate that the machine learning system is more appropriate than the rule-based system for providing optimal adjustment of the radio, where the underlying mechanisms are too complex to allow simple rules of thumb to be applied. However, the rule-based technique has been shown to be suitable for areas with large nonlinearities. A combination of the two techniques would solve some of the problems caused by the suitability of each method to a specific problem type, the rule-based approach covering the areas with large nonlinearities and the machine learning system the more linear regions. This would help in simplifying the implementation as it would minimize the number of rules required for each different radio type.>