Progress beyond the State of the Art

The principal objective of the project is to advance the state-of-the-art in versatile active perception systems of a kind suitable for deployment in cognitive robotic systems. The state-of-the-art at the start of the project is summarised below

Shape:

Full 3D structure can be obtained for static objects using micro-CT scanning. Unfortunately, the process takes several hours and cannot be applied to living bats. However, dense correlation stereo devices capable of the resolution needed are newly available.

We shall develop the dense correlation stereo vision technique further to allow high-resolution capture of dynamic shape, by first producing reduced parameter models from static and nearly static bat heads, and then fitting these models to the dynamic data.

Much of the necessary work is data capture and model fitting. This will require state-of-the-art engineering and data analysis, but is not expected to require much innovation. On the other hand, high speed 3D capture of objects in flight and development of a low dimensional deformable surface model from the noisy 3D video data are distinctly novel and will be easily publishable. At the moment we are not aware of any groups worldwide even having equipment that could capture such data.

Acoustics

Acoustic simulation tools based on finite element modelling principles, optimised for the problem of computing acoustic beam directivities from volumetric models of ear and noseleaf shapes --- while allowing investigation of which features of the shape contribute to which features of the directivity --- were developed during the CIRCE project and have subsequently been enhanced and further optimised.

UA has adapted boundary element modelling techniques to model the directivity of a complete bat head given a scanned model of the shape. Similar boundary element methods are employed by the Acoustic Simulation group at SDU, though for lower frequency and larger structure applications.

While the various methods above are able to simulate the beamforming effects of the static morphology of bats, they fail to address two key issues relevant to this project: first, the simulation of deformable structures whose shape may change during the emission and reception of acoustic signals; and second, the methods have not been fully validated.

Systems

The most relevant complete artificial system available is the CIRCE biomimetic robotic bat head. This is a complete biosonar model in a realistic physical size; however no attempt was given to modelling realistic behaviours.

Living bats are complete systems which have the abilities we wish to emulate, but the information required for engineering them is woefully limited, with only partial knowledge of acoustic features and little attention paid to closed-loop behaviour.

During this project we will collect the specific data needed to make an accurate reconstruction of the bats' acoustic experiences, sufficient to build physical models. The reconstructed acoustic experience will reveal cues bats may use to direct their behaviour. We shall also refine the inferences drawn by feeding results from our models back into real bat experiments, and repeating this spiral design process as long as time permits.

As an essential part of the pursuit of the research programme, new techniques for shape and motion capture of bats will be developed, new sonar-related techniques --- for example for beamforming --- will be developed, existing acoustic simulation techniques will be extended, and considerable new knowledge relating to the abilities and behaviour of the studied bat species will be produced, which will be employed in two versatile and robust active sonar systems suitable for application to detection and classification tasks in the robotics domain.

Anticipated State-of-the-Art on Completion

With respect to the foregoing description of the current state-of-the-art relevant to the proposed project, we anticipate that the following advances will have been made once the ChiRoPing research programme is complete:

• Two complete target detection, localisation and capture systems exploiting the acoustic and movement options available to a bat-like active sonar system will have been constructed, characterised and evaluated from both engineering and biological perspectives.
• Novel techniques for the acquisition of small, complex convoluted shapes using computer vision will have been developed, demonstrated and their performance quantified with respect to volumetric scanning methods such as micro-CT.
• Acoustic modelling techniques suitable for small deformable and deforming thin convoluted shapes and ultrasonic frequencies, and quantification of their performance, will be available.
• A database of bat hunting behaviour instances --- comprising flight trajectory, reconstructed acoustic emissions, position and morphology of head, ears and noseleaf annotated with behaviour interpretation for each bat species studied in the project --- will be available to biologists and computational modellers interested in bat behaviour.
• Biomimetic sonar hardware based closely on the studied bats, including dynamically variable beamforming technology, will be available to roboticists and other interested researchers.
• Computational models of the hunting behaviours of two or more species of bat will be available for bat biologists to study, resulting in novel research questions, deeper understanding of the details of bat biosonar, and more complete knowledge of the specific bat species studied here.