Circuits - Computation - Models: Engineering

Engineering

<div style="text-align: justify;"><strong>Autonomous quadrocoptor using visual orientation based on lobula plate cell network.<br /> a</strong>. Image of the quadrocopter. <strong>b</strong> The eye of the quadrocoptor is a small CCD camera with a fish eye lense. <strong>c</strong> Example image during roll and corresponding optic field as calculated from the artificial LPTC network.</div> Zoom Image
Autonomous quadrocoptor using visual orientation based on lobula plate cell network.
a
. Image of the quadrocopter. b The eye of the quadrocoptor is a small CCD camera with a fish eye lense. c Example image during roll and corresponding optic field as calculated from the artificial LPTC network.
[less]

Our knowledge about the fly motion vision system goes directly into the development of stabilization and guidance systems for miniature airborne robots. The aim is to build a fully autonomous aerial vehicle capable of avoiding obstacles through the same optic flow techniques that a fly employs to navigate through its world. The principles of the flies’ optic flow processing lend themselves well for technical systems due to their robustness and parallel way of computation: despite possessing exceedingly small numbers of neurons compared to higher organisms flies are exceptionally skilled at avoiding obstacles - even at very high speeds. This is all due to their highly optimized visual system. Building such a system generates a lot of technical know-how and insight into what problems arise and need to be solved for this type of autonomous robot to fly in a controlled manner. The hope is that we can also in turn learn from this exercise to better understand the intricate neural circuitry we find in the fly visual processing machine.

We have therefore built a fly-inspired visual sensor system by implementing the fly’s approach of optic flow processing on board a custom built high-speed digital processing platform (Plett et al., 2012). This light-weight sensor system was then incorporated into a miniature quadrocopter, which is capable of hovering as well as rapid flight maneuvers. After successfully achieving fully autonomous flight we are now seeking to implement this fly-inspired sensing scheme in a fully analog sensor design for further miniaturization and optimization.

This work is done in collaboration with Martin Buss and Kolja Kuehnlenz from the Technische Universität München (TUM) and sponsored by the BMBF within the excellence cluster CoTeSys.

 

References

Plett, J., Bahl, A., Buss, M., Kühnlenz, K. & Borst, A. Bio-inspired visual ego-rotation sensor for MAVs. Biol. Cybern. 106, 51–63 (2012).

 
loading content