The blowfly Calliphora is one of the most sophisticated fliers in the animal kingdom.
It displays a broad repertoire of visually guided behaviours that can readily
be quantified, including gaze and flight stabilization reflexes, male chasing flights,
collision avoidance and landing responses. The fly achieves such robust visuo-motor
control tasks based on a comparatively simple nervous system that is highly
accessible for electrophysiological recordings. The ability to investigate the fly’s
performance at the behavioural and electrophysiology levels makes this animal an
ideal model system to study closed-loop visual motor control.
The aim of this thesis was to develop and characterize the dynamics of a fly-robot
interface (FRI) while a fly performs a closed-loop visual stabilization task.
A novel experimental setup involving a FRI was developed which allowed for simultaneous
measurements of neural activity from the fly and the behavioural performance
of the robot. In the setup, the neural activity of an identified visual interneuron,
the H1 cell, was recorded and its action potentials were used to control
the motion of a mobile robot that was free to rotate along its vertical axis. External
visual perturbations were introduced into the closed-loop system through a rotating
turn-table with the robot using the neural activity to counter-rotate and to minimize
the observed visual motion. The closed-loop control delay of the FRI was 50 ms
which is well within the range of visual response delays observed in fly behaviour.
With the FRI, the closed-loop dynamics of a static-gain proportional controller
were characterized. The results explain significant oscillations in the closed-loop
responses as a possible consequence of a high controller gain which were also
observed but never fully interpreted in previous behavioural studies. Varying the controller gain also offers competing control benefits to the fly, with different gains
maximizing performance for different input frequency ranges and thus different
behavioural tasks.
Results with the proportional controller indicate the dependence of the FRI frequency
response on the angular acceleration of visual motion. An adaptive controller
designed to dynamically scale the feedback gain was found to increase the
bandwidth of the frequency response when compared with the static-gain proportional
controller.
The image velocities observed under closed-loop conditions using the proportional
and the adaptive controllers were correlated with the spiking activity of the H1-cell.
A remarkable qualitative similarity was found between the response dynamics of
the cell under closed-loop conditions with those obtained in previous open-loop
experiments. Specifically, (i) the peak spike rate decreased when the mean image
velocity was increased, (ii) the relationship between spike rate and image velocity
was dependant on the standard deviation of the image velocities suggesting adaptive
scaling of the cell’s signalling range, and (iii) the cell’s gain decreased linearly
with increasing image accelerations.
Despite the fact that several sensory modalities - including the motion vision pathway
- process information in a non-linear fashion signal integration at stages one to
two synapses away from the motor systems and the behavioural output itself have
been shown to be linear. Quantifying the closed-loop dynamics of visuo-motor
control at both the behavioural and neuronal level, may provide a starting point to
discover the neural mechanisms underlying an appropriate combination of complementary
non-linear processes which ultimately result in a linear performance of the
overall system.