Concept

The principal objective of the project is to discover ways of realising robust and versatile embodied active sonar sensor systems based on bats, who are Nature's experts in this technology. Such sensors complement visual sensing and, as attested by bats, offer sufficient discriminatory ability to support new kinds of cognitive robotic systems.

Versatile and robust systems able to respond sensibly to challenges not precisely specified in their design depend on the appropriate integration of at least the following factors:

• specific morphology, by means of which signal processing such as directionality or antenna gain may be implemented directly in hardware;
• specific behaviours associated with the sensor, through which the environment is sampled and constrained in a task-dependent manner;
• for an active sensor, the structure of signals emitted, which constrains the signals returned from the environment and can be used to highlight perceptually salient characteristics while diminishing or eliminating distractors.

Bats exploit all three components in an integrated and coherent solution through which they realise tangible object discrimination and handling in real time with outstanding success. Unfortunately, very little is known about exactly how bats deploy these components to achieve their robust and versatile hunting behaviours in their varied habitats; without such knowledge, engineering equivalent sonar systems is impossible. The goal of this project is to remedy that lack of knowledge.

The strategy we propose for accomplishing this goal is one of biomimesis in its classic form --- that is, the construction and evaluation of working models of specific biological systems using engineering techniques and predicated on a close collaboration between experts in bat ethology and in robotics.

To maintain a focus on the engineering goal of generic robust and versatile sensory discrimination in realistically complex environments we propose to model two specific and contrasting bat hunting behaviours. First, we shall model a gleaning bat which picks stationary insects from leaves, in the hope of finding its mechanism of discovering interesting cues in the presence of many distracting ones. Second, we shall model the ability of a trawling bat to capture prey on water surfaces without either colliding with the water or attacking non-prey clutter objects floating on the water surface. We have chosen three taxonomically distant bat species that all hunt over water, in order to determine the shared adaptations and thus generalise what is important for a sonar sensing system operating in this habitat. These three trawling bats show striking differences in the sound emitting and receiving features, i.e., mouth, face and ears, as well as in the structure of their echolocation calls, so a wide range of sonar options can be studied. We shall make only one trawling bat model, as a testcase of our ability to extract general design rules for active perception systems.

The primary evaluation of the two biomimetic systems will be based on engineering criteria: robustness, versatility, and ease of application. We shall, however, limit ourselves to tasks equivalent to those the bats routinely perform, to reduce the time needed. The systems will also be evaluated from a biologist's standpoint, in terms of how well they predict the failure modes and previously unmeasured abilities of their living prototypes.

To ensure the fidelity of the models constructed to their original inspirations --- a requirement motivated by both engineering and biological reasons --- we shall conduct appropriate behavioural experiments with bats of various species under semi-natural conditions. From these experiments we shall elicit, for the first time, a comprehensive set of synchronised flight, vocalisation, morphology and behaviour data for the species studied, which will be used to derive the models.

Measurable Objectives

To summarise, the project will develop two embodied active sonar perception systems, which are biomimetic models of the bats mentioned above, demonstrate them in tasks that require handling of %a variety of tangible objects to be detected among clutter, and evaluate them thoroughly from both an engineering and a biological standpoint.

This gives the following measurable objectives:

1. a database of acoustic, morphological and behavioural data for bats engaged in natural hunting behaviours;
2. two computational models of the chosen bats engaged in natural hunting behaviours, implemented as engineered sensor systems;
3. an evaluation of the resulting sensor systems both as engineered devices applied in realistic engineering contexts and as biological models of their respective bat species.