Final Report Summary - CHIROCOPTER (A remote controlled helicopter for investigating the echoes experienced by bat during navigation)
Background
While foraging and commuting, echolocating bats navigate to salient places in their surroundings. They are able to do this relying only on their sense of echolocation. Currently, we have only a very limited idea about the kind of information these animals extract from the returning echoes to support navigation.
To better understand how bats use echoes to recognize places and navigate to them it is essential to know the sensorial input encountered by bats as they fly through natural environments. So far, only a few studies have collected and analysed the echoes generated by a small number of reflectors encountered by bats such as prey items and vegetation (recently reviewed by Yovel et al., 2011). However, no data exists on the statistics of the echoes generated by the complex environments encountered by bats during flight. In other words, we know what the echoes from a plant look like. However, we do know how the echoes of a row of vegetation changes as the bat flies past it. This implies that it is unknown how the echoes received by bats change from call to call and which information is available in this flow of changing echoes.
This project aimed at collecting a large sample of echoes from the environments as they would be experienced by bats flying through those environments. To do this, an ensonification device, consisting of an ultrasonic emitter and an array of microphones, was built. This device was used to scan a number of outdoor environments and collect a larger sample of echoes. Over 2 million pulse-echo pairs were collected.
A movie of the ensonification device being used to scan parts of a city park can be found at https://www.youtube.com/watch?v=HeyAXrP-qYI(opens in new window)
In this approach, this project is the first to collect an ecologically valid set of stimuli (echoes) as perceived by flying bats. The dataset thus collected has allowed testing hypothesises about the navigation abilities of bats. In particular, we proposed a model of bat navigation. In relating work, we have proposed how the results of this project can be used to enhance the way in which robots used echolocation.
Model of bat navigation
Currently, it is assumed (often implicitly) that bat reconstruct a 3D layout of the environment. They are assumed to use this reconstruction to recognize places and navigate to them. However, evidence for bats being able to reconstruct a 3D layout of a scene is sparse. Indeed, a number of limitations of sonar render it unlikely that bats can derive a 3D model of the world from echoes.
We propose an alternative way bats may identify previously visited places using sonar. The template based identification approach we propose does not assume bats reconstruct a representation of the environment. In contrast, we suggest that the cochlear output can be used directly to identify locations. This is, instead of attempting to reconstruct the position, shape and identity of objects from the echoes, we propose that the echoes might be used directly quasi directly. Under this hypothesis, bats are assumed to match the output of the cochlea directly to a set of stored templates without deriving a model or reconstructing the layout of the reflectors. Simply stated, bats would listen to how an environment sounds rather than to try and infer how it looks (i.e. deriving the 3D layout of objects).
Using a template matching approach, bats would avoid the need for complex reconstruction algorithms extracting 3D information from a 1D signal at each ear. We propose that a bat only has to recognize the echo signature of a scene to identify it. It does not need to reconstruct the 3D structure. In addition to circumventing the computationally hard problem of deriving a 3D spatial representation from complex echoes, acoustic templates have been shown to be very discriminative. For example, Roman Kuc (1997) showed a bio-mimetic sonar device to be able to detect which side of a coin was showing using very simple templates.
In this project, using ensonification data, we derived templates from a large number of echo trains from natural habitats. Next, we assessed the viability of our alternative hypothesis by evaluating whether the derived templates are sufficient to support scene recognition and navigation. Using a number of mathematical measures to evaluate the properties of the templates, we tentatively concluded that the templates could support navigation. In addition, simulation studies have been carried out showing how the bat could use the templates to build a map of the environment (e.g. a forest). To the best of our knowledge, our work is the first explicit (and quantitaive) model of how bats could navigate using sonar.
The insights gained from the project also have an impact in engineering. Sonar sensors are found on many robots. Despite this ubiquity of sonar sensors, the way in which robots use sonar is very impoverished when compared to biological echolocators such as bats. To date, robots use sonar mainly as a means of estimating distances or detecting obstacles using time of flight. In contrast, bats can navigate and forage under very demanding circumstances relying solely on sonar. Therefore, it is unsurprising that some research has aimed at mimicking bat sonar in robots. However, this has not resulted in design principles to guide the implementation of bio-inspired sonar on robots. Drawing on the findings of this projects, we derived a set of design principles for bio-inspired sonar. In the paper describing this we, propose a framework for implementing bio-inspired sonar along these principles. Moreover, we demonstrate how a bio-inspired navigation strategy could be implemented. We conclude that the principles and architectural framework presented here are a step towards substantially expanding the range of applications for sonar in robotics.
Key references
Yovel, Y., Franz, M. O., Stilz, P., & Schnitzler, H. U. (2011). Complex echo classification by echo-locating bats: a review. Journal of Comparative Physiology A, 197(5), 475-490.
Kuc, R. B. (1997). Biomimetic sonar differentiates coin head from tail. The Journal of the Acoustical Society of America, 101(5), 3198-3198.
While foraging and commuting, echolocating bats navigate to salient places in their surroundings. They are able to do this relying only on their sense of echolocation. Currently, we have only a very limited idea about the kind of information these animals extract from the returning echoes to support navigation.
To better understand how bats use echoes to recognize places and navigate to them it is essential to know the sensorial input encountered by bats as they fly through natural environments. So far, only a few studies have collected and analysed the echoes generated by a small number of reflectors encountered by bats such as prey items and vegetation (recently reviewed by Yovel et al., 2011). However, no data exists on the statistics of the echoes generated by the complex environments encountered by bats during flight. In other words, we know what the echoes from a plant look like. However, we do know how the echoes of a row of vegetation changes as the bat flies past it. This implies that it is unknown how the echoes received by bats change from call to call and which information is available in this flow of changing echoes.
This project aimed at collecting a large sample of echoes from the environments as they would be experienced by bats flying through those environments. To do this, an ensonification device, consisting of an ultrasonic emitter and an array of microphones, was built. This device was used to scan a number of outdoor environments and collect a larger sample of echoes. Over 2 million pulse-echo pairs were collected.
A movie of the ensonification device being used to scan parts of a city park can be found at https://www.youtube.com/watch?v=HeyAXrP-qYI(opens in new window)
In this approach, this project is the first to collect an ecologically valid set of stimuli (echoes) as perceived by flying bats. The dataset thus collected has allowed testing hypothesises about the navigation abilities of bats. In particular, we proposed a model of bat navigation. In relating work, we have proposed how the results of this project can be used to enhance the way in which robots used echolocation.
Model of bat navigation
Currently, it is assumed (often implicitly) that bat reconstruct a 3D layout of the environment. They are assumed to use this reconstruction to recognize places and navigate to them. However, evidence for bats being able to reconstruct a 3D layout of a scene is sparse. Indeed, a number of limitations of sonar render it unlikely that bats can derive a 3D model of the world from echoes.
We propose an alternative way bats may identify previously visited places using sonar. The template based identification approach we propose does not assume bats reconstruct a representation of the environment. In contrast, we suggest that the cochlear output can be used directly to identify locations. This is, instead of attempting to reconstruct the position, shape and identity of objects from the echoes, we propose that the echoes might be used directly quasi directly. Under this hypothesis, bats are assumed to match the output of the cochlea directly to a set of stored templates without deriving a model or reconstructing the layout of the reflectors. Simply stated, bats would listen to how an environment sounds rather than to try and infer how it looks (i.e. deriving the 3D layout of objects).
Using a template matching approach, bats would avoid the need for complex reconstruction algorithms extracting 3D information from a 1D signal at each ear. We propose that a bat only has to recognize the echo signature of a scene to identify it. It does not need to reconstruct the 3D structure. In addition to circumventing the computationally hard problem of deriving a 3D spatial representation from complex echoes, acoustic templates have been shown to be very discriminative. For example, Roman Kuc (1997) showed a bio-mimetic sonar device to be able to detect which side of a coin was showing using very simple templates.
In this project, using ensonification data, we derived templates from a large number of echo trains from natural habitats. Next, we assessed the viability of our alternative hypothesis by evaluating whether the derived templates are sufficient to support scene recognition and navigation. Using a number of mathematical measures to evaluate the properties of the templates, we tentatively concluded that the templates could support navigation. In addition, simulation studies have been carried out showing how the bat could use the templates to build a map of the environment (e.g. a forest). To the best of our knowledge, our work is the first explicit (and quantitaive) model of how bats could navigate using sonar.
The insights gained from the project also have an impact in engineering. Sonar sensors are found on many robots. Despite this ubiquity of sonar sensors, the way in which robots use sonar is very impoverished when compared to biological echolocators such as bats. To date, robots use sonar mainly as a means of estimating distances or detecting obstacles using time of flight. In contrast, bats can navigate and forage under very demanding circumstances relying solely on sonar. Therefore, it is unsurprising that some research has aimed at mimicking bat sonar in robots. However, this has not resulted in design principles to guide the implementation of bio-inspired sonar on robots. Drawing on the findings of this projects, we derived a set of design principles for bio-inspired sonar. In the paper describing this we, propose a framework for implementing bio-inspired sonar along these principles. Moreover, we demonstrate how a bio-inspired navigation strategy could be implemented. We conclude that the principles and architectural framework presented here are a step towards substantially expanding the range of applications for sonar in robotics.
Key references
Yovel, Y., Franz, M. O., Stilz, P., & Schnitzler, H. U. (2011). Complex echo classification by echo-locating bats: a review. Journal of Comparative Physiology A, 197(5), 475-490.
Kuc, R. B. (1997). Biomimetic sonar differentiates coin head from tail. The Journal of the Acoustical Society of America, 101(5), 3198-3198.