Improvisation in interactive music systems

This project investigates machine listening and improvisation in interactive music systems with the goal of improvising musically appropriate accompaniment to an audio stream in real-time. The input audio may be from a live musical ensemble, or playback of a recording for use by a DJ. I present a collection of robust techniques for machine listening in the context of Western popular dance music genres, and strategies of improvisation to allow for intuitive and musically salient interaction in live performance. The findings are embodied in a computational agent – the Jambot – capable of real-time musical improvisation in an ensemble setting. Conceptually the agent’s functionality is split into three domains: reception, analysis and generation. The project has resulted in novel techniques for addressing a range of issues in each of these domains. In the reception domain I present a novel suite of onset detection algorithms for real-time detection and classification of percussive onsets. This suite achieves reasonable discrimination between the kick, snare and hi-hat attacks of a standard drum-kit, with sufficiently low-latency to allow perceptually simultaneous triggering of accompaniment notes. The onset detection algorithms are designed to operate in the context of complex polyphonic audio. In the analysis domain I present novel beat-tracking and metre-induction algorithms that operate in real-time and are responsive to change in a live setting. I also present a novel analytic model of rhythm, based on musically salient features. This model informs the generation process, affording intuitive parametric control and allowing for the creation of a broad range of interesting rhythms. In the generation domain I present a novel improvisatory architecture drawing on theories of music perception, which provides a mechanism for the real-time generation of complementary accompaniment in an ensemble setting. All of these innovations have been combined into a computational agent – the Jambot, which is capable of producing improvised percussive musical accompaniment to an audio stream in real-time. I situate the architectural philosophy of the Jambot within contemporary debate regarding the nature of cognition and artificial intelligence, and argue for an approach to algorithmic improvisation that privileges the minimisation of cognitive dissonance in human-computer interaction. This thesis contains extensive written discussions of the Jambot and its component algorithms, along with some comparative analyses of aspects of its operation and aesthetic evaluations of its output. The accompanying CD contains the Jambot software, along with video documentation of experiments and performances conducted during the project.

CAT-SLAM : probabilistic localisation and mapping using a continuous appearance-based trajectory

This paper describes a new system, dubbed Continuous Appearance-based Trajectory Simultaneous Localisation and Mapping (CAT-SLAM), which augments sequential appearance-based place recognition with local metric pose filtering to improve the frequency and reliability of appearance-based loop closure. As in other approaches to appearance-based mapping, loop closure is performed without calculating global feature geometry or performing 3D map construction. Loop-closure filtering uses a probabilistic distribution of possible loop closures along the robot’s previous trajectory, which is represented by a linked list of previously visited locations linked by odometric information. Sequential appearance-based place recognition and local metric pose filtering are evaluated simultaneously using a Rao–Blackwellised particle filter, which weights particles based on appearance matching over sequential frames and the similarity of robot motion along the trajectory. The particle filter explicitly models both the likelihood of revisiting previous locations and exploring new locations. A modified resampling scheme counters particle deprivation and allows loop-closure updates to be performed in constant time for a given environment. We compare the performance of CAT-SLAM with FAB-MAP (a state-of-the-art appearance-only SLAM algorithm) using multiple real-world datasets, demonstrating an increase in the number of correct loop closures detected by CAT-SLAM.

Improved facial expression recognition via uni-hyperplane classification

Large margin learning approaches, such as support vector machines (SVM), have been successfully applied to numerous classification tasks, especially for automatic facial expression recognition. The risk of such approaches however, is their sensitivity to large margin losses due to the influence from noisy training examples and outliers which is a common problem in the area of affective computing (i.e., manual coding at the frame level is tedious so coarse labels are normally assigned). In this paper, we leverage the relaxation of the parallel-hyperplanes constraint and propose the use of modified correlation filters (MCF). The MCF is similar in spirit to SVMs and correlation filters, but with the key difference of optimizing only a single hyperplane. We demonstrate the superiority of MCF over current techniques on a battery of experiments.

A requirements framework for semantic business process modelling

The emergence of semantic technologies to deal with the underlying meaning of things, instead of a purely syntactical representation, has led to new developments in various fields, including business process modeling. Inspired by artificial intelligence research, technologies for semantic Web services have been proposed and extended to process modeling. However, the applicablility of semantic Web services for semantic business processes is limited because business processes encompass wider requirements of business than Web services. In particular, processes are concerned with the composition of tasks, that is, in which order activities are carried out, regardless of their implementation details; resources assigned to carry out tasks, such as machinery, people, and goods; data exchange; and security and compliance concerns.

Force controlled hexapod walking

In this paper we describe the dynamic simulation of an 18 degrees of freedom hexapod robot with the objective of developing control algorithms for smooth, efficient and robust walking in irregular terrain. This is to be achieved by using force sensors in addition to the conventional joint angle sensors as proprioceptors. The reaction forces on the feet of the robot provide the necessary information on the robots interaction with the terrain. As a first step we validate the simulator by implementing movement control by joint torques using PID controllers. As an unexpected by-product we find that it is simple to achieve robust walking behaviour on even terrain for a hexapod with the help of PID controllers and by specifying a trajectory of only a few joint configurations.

ACE research briefing paper 002: learning and adaptation in the business creation process

This year marks the completion of data collection for year three (Wave 3) of the CAUSEE study. This report uses data from the first three years and focuses on the process of learning and adaptation in the business creation process. Most start-ups need to change their business model, their product, their marketing plan, their market or something else about the business to be successful. PayPal changed their product at least five times, moving from handheld security, to enterprise apps, to consumer apps, to a digital wallet, to payments between handhelds before finally stumbling on the model that made the a multi-billion dollar company revolving around email-based payments. PayPal is not alone and anecdotes abounds of start-ups changing direction: Sysmantec started as an artificial intelligence company, Apple started selling plans to build computers and Microsoft tried to peddle compilers before licensing an operating system out of New Mexico. To what extent do Australian new ventures change and adapt as their ideas and business develop? As a longitudinal study, CAUSEE was designed specifically to observe development in the venture creation process. In this research briefing paper, we compare development over time of randomly sampled Nascent Firms (NF) and Young Firms(YF), concentrating on the surviving cases. We also compare NFs with YFs at each yearly interval. The 'high potential' over sample is not used in this report.

Outdoor sensornet design and deployment : experiences from a sugar farm

A wireless sensor network collected real-time water-quality measurements to investigate how current irrigation practices—in particular, underground water salination—affect the environment. New protocols provided high end-to-end packet delivery rates in the hostile deployment environment.

Flight trial of an electro-optical sense-and-avoid system

This paper presents the flight trials of an electro-optical (EO) sense-and-avoid system onboard a Cessna host aircraft (camera aircraft). We focus on the autonomous collision avoidance capability of the sense-and-avoid system; that is, closed-loop integration with the onboard aircraft autopilot. We also discuss the system’s approach to target detection and avoidance control, as well as the methodology of the flight trials. The results demonstrate the ability of the sense-and-avoid system to automatically detect potential conflicting aircraft and engage the host Cessna autopilot to perform an avoidance manoeuvre, all without any human intervention

Hessian-based affine adaptation of salient local image features

Affine covariant local image features are a powerful tool for many applications, including matching and calibrating wide baseline images. Local feature extractors that use a saliency map to locate features require adaptation processes in order to extract affine covariant features. The most effective extractors make use of the second moment matrix (SMM) to iteratively estimate the affine shape of local image regions. This paper shows that the Hessian matrix can be used to estimate local affine shape in a similar fashion to the SMM. The Hessian matrix requires significantly less computation effort than the SMM, allowing more efficient affine adaptation. Experimental results indicate that using the Hessian matrix in conjunction with a feature extractor that selects features in regions with high second order gradients delivers equivalent quality correspondences in less than 17% of the processing time, compared to the same extractor using the SMM.

Activity analysis in complicated scenes using DFT coefficients of particle trajectories

Modelling activities in crowded scenes is very challenging as object tracking is not robust in complicated scenes and optical flow does not capture long range motion. We propose a novel approach to analyse activities in crowded scenes using a “bag of particle trajectories”. Particle trajectories are extracted from foreground regions within short video clips using particle video, which estimates long range motion in contrast to optical flow which is only concerned with inter-frame motion. Our applications include temporal video segmentation and anomaly detection, and we perform our evaluation on several real-world datasets containing complicated scenes. We show that our approaches achieve state-of-the-art performance for both tasks.

Unusual scene detection using distributed behaviour model and sparse representation

The ability to detect unusual events in surviellance footage as they happen is a highly desireable feature for a surveillance system. However, this problem remains challenging in crowded scenes due to occlusions and the clustering of people. In this paper, we propose using the Distributed Behavior Model (DBM), which has been widely used in computer graphics, for video event detection. Our approach does not rely on object tracking, and is robust to camera movements. We use sparse coding for classification, and test our approach on various datasets. Our proposed approach outperforms a state-of-the-art work which uses the social force model and Latent Dirichlet Allocation.

Visible and near infrared spectroscopy of Hayabusa re-entry using semi-autonomous tracking

A ground-based tracking camera and co-aligned slit-less spectrograph were used to measure the spectral signature of visible radiation emitted from the Hayabusa capsule as it entered into the Earth's atmosphere in June 2010. Good quality spectra were obtained that showed the presence of radiation from the heat shield of the vehicle and the shock-heated air in front of the vehicle. An analysis of the black body nature of the radiation concluded that the peak average temperature of the surface was about (3100±100) K.

Beyond here-and-now : extending shared physical experiences to shared conceptual experiences

For robots to use language effectively, they need to refer to combinations of existing concepts, as well as concepts that have been directly experienced. In this paper, we introduce the term generative grounding to refer to the establishment of shared meaning for concepts referred to using relational terms. We investigated a spatial domain, which is both experienced and constructed using mobile robots with cognitive maps. The robots, called Lingodroids, established lexicons for locations, distances, and directions through structured conversations called where-are-we, how-far, what-direction, and where-is-there conversations. Distributed concept construction methods were used to create flexible concepts, based on a data structure called a distributed lexicon table. The lexicon was extended from words for locations, termed toponyms, to words for the relational terms of distances and directions. New toponyms were then learned using these relational operators. Effective grounding was tested by using the new toponyms as targets for go-to games, in which the robots independently navigated to named locations. The studies demonstrate how meanings can be extended from grounding in shared physical experiences to grounding in constructed cognitive experiences, giving the robots a language that refers to their direct experiences, and to constructed worlds that are beyond the here-and-now.

Visual route recognition with a handful of bits

In this paper we use a sequence-based visual localization algorithm to reveal surprising answers to the question, how much visual information is actually needed to conduct effective navigation? The algorithm actively searches for the best local image matches within a sliding window of short route segments or 'sub-routes', and matches sub-routes by searching for coherent sequences of local image matches. In contract to many existing techniques, the technique requires no pre-training or camera parameter calibration. We compare the algorithm's performance to the state-of-the-art FAB-MAP 2.0 algorithm on a 70 km benchmark dataset. Performance matches or exceeds the state of the art feature-based localization technique using images as small as 4 pixels, fields of view reduced by a factor of 250, and pixel bit depths reduced to 2 bits. We present further results demonstrating the system localizing in an office environment with near 100% precision using two 7 bit Lego light sensors, as well as using 16 and 32 pixel images from a motorbike race and a mountain rally car stage. By demonstrating how little image information is required to achieve localization along a route, we hope to stimulate future 'low fidelity' approaches to visual navigation that complement probabilistic feature-based techniques.

Towards persistent localization and mapping with a continuous appearance-based topology

Appearance-based localization can provide loop closure detection at vast scales regardless of accumulated metric error. However, the computation time and memory requirements of current appearance-based methods scale not only with the size of the environment but also with the operation time of the platform. Additionally, repeated visits to locations will develop multiple competing representations, which will reduce recall performance over time. These properties impose severe restrictions on long-term autonomy for mobile robots, as loop closure performance will inevitably degrade with increased operation time. In this paper we present a graphical extension to CAT-SLAM, a particle filter-based algorithm for appearance-based localization and mapping, to provide constant computation and memory requirements over time and minimal degradation of recall performance during repeated visits to locations. We demonstrate loop closure detection in a large urban environment with capped computation time and memory requirements and performance exceeding previous appearance-based methods by a factor of 2. We discuss the limitations of the algorithm with respect to environment size, appearance change over time and applications in topological planning and navigation for long-term robot operation.