Contents
I am engaged in a number of projects in biology and engineering. As time permits, I post interesting images or movies from my work here.
Vision and control of flight
A 3D fly trajectory captured as a fly explored its environment, a panoramic background texture on the inner wall of a 2 meter diameter cylinder and four vertical posts.
My primary research project concerns the control of flight of the fruit fly, Drosophila melanogaster. While we understand much about visual guidance of behavior from focused laboratory experiments in which a single behavior or class of neurons was investigated, we know much less about how these behaviors and neural responses are combined to give rise to a repertoire of behaviors that allow flies to successfully avoid dangerous situations, to find food, to find mates, and ultimately to reproduce. To address these issues, I am studying how the behavior of a freely flying fly corresponds to well-characterized behaviors from experiments in more restricted conditions. Of particular interest is high-level control. Is fly behavior an emergent property of multiple low-level control laws, or do top-down processes govern switching betwen sub-behaviors? Can overall patterns of behavior be explained by switching between identified behaviors or is the overall pattern of behavior better explained by a continuously ongoing summation of low-level reflexes?
To address these issues, I am performing experiments in which I track the location of freely flying flies in a large arena (2m diameter) while presenting a variety of visual and olfactory stimuli. This required the construction of a multiple camera fly tracking system that is capable of resolving the position of flies over a large volume.
The multiple camera fly tracking system works with arbitrary numbers of cameras (8 shown of 11 total for this experiment) to track the position of a freely flying fly in a large environment.
A fly's eye view of the world
To facilitate studies of visually guided behavior in the fly, I created a software package to model fly eye in a biologically realistic way. This model integrates the findings from many experimental studies by many authors and simulates the optical and neural processing occuring in Drosophila vision. Part of this model, the Drosophila eye map is available to anyone, while the rest is part of the Dickinson Lab's GUF model.
A reconstruction of the view experienced by a fly as it navigated around the arena described above. Note that azimuth angle is assumed to be in the direction of travel and that elevation and bank angles are fixed.
Together with Will Dickson and Michael Dickinson, we used this model visual system to provide the input to a control system that steered a flapping model fly through a virtual environment. We published this work in:
Dickson, W.B., Straw, Andrew D., Dickinson, M.H. (2008) Integrative Model of Drosophila Flight. AIAA Journal 46(9). doi: 10.2514/1.29862
An inexpensive testbed for insect inspired control
We (Shuo Han, myself, Michael Dickinson, and Richard Murray) have created an inexpensive platform to test insect-inspired control on a conventional flight platform. The basic idea is that a helicopter guidance system sees through optics and a motion detection system similar to that of the fruit fly, and we program a computer to control flight based on that input. The key to keeping our platform inexpensive is that that, rather than have onboard cameras and optics, we track the helicopter in realtime and simulate what a fly would see by immersing our helicopter in a simulated 3D environment created with a computer games engine. A paper describing this work in more detail has been submitted for publication:
Han, Shuo, Straw, Andrew D., Dickinson, Michael H. & Murray, Richard M. (submitted Proceedings of the 2009 IEEE Conference on Robotics and Automation) A Real-Time Helicopter Testbed for Insect-Inspired Visual Flight Control.
An inexpensive toy helicopter is flying under autonomous control from a computer (shown in the upper panel). A $200 camera under the piece of glass looks upwards and localizes several markers on the underside of the helicopter in a known configuration (using the trackem component of my motmot software). With the standard POSIT algorithm, we calculate the position and orientation of the helicopter within about 25 millseconds and then simulate the fly eye view (in the lower panel). The output of simulated Hassenstein-Reichardt correlator motion detectors is then used as input for a yaw controller. The commanded yaw is 30 degrees/second for the first 40 seconds, then 0 degrees/second. (Here, translational position is stabilized using state feedback estimated directly from the POSIT algorithm.)
Page last modified Wed Oct 8 23:43:00 2008.