Lecoeur Julien, Dacke Marie, Floreano Dario, Baird Emily (2019), The role of optic flow pooling in insect flight control in cluttered environments, in Scientific Reports
, 9(1), 7707-7707.
Schilling Fabian, Lecoeur Julien, Schiano Fabrizio, Floreano Dario (2019), Learning Vision-Based Flight in Drone Swarms by Imitation, in IEEE Robotics and Automation Letters
, 4(4), 4523-4530.
Lecoeur Julien, Baird Emily, Floreano Dario (2018), Spatial Encoding of Translational Optic Flow in Planar Scenes by Elementary Motion Detector Arrays, in Scientific Reports, Springer US
, 8(1), 5821.
We will develop novel vision-based controllers for mini-drones handle small events in static and dynamic environments. The project is divided in two parts. In the first part, we will address flight control in complex static environments made of a random layout of several objects at different distances from the agent, which capture typical situations while flying through a forest of trees. This part will involve the modeling of vision-based strategies used by flying insects to maneuver safely among obstacles. These strategies will be implemented on autonomous microflyers and characterized in environments where the navigation goal cannot be reached by state-of-the-art algorithms, such as flying through narrow openings against a richly textured environment. In the second part, we will address strategies to deal with small dynamic events, such as other agents (insects or robots) flying in the same environment, in order to avoid, follow, and fly in formation with them. We will take inspiration from the biological evidence that specifically tuned and strategically located receptive fields could be responsible for the detection and tracking of other moving objects in the visual field.We will benefit from an ongoing cooperation with Prof. Emily Baird in the Biology Department of the University of Lund in Sweden, who will carry out experimental work with real insects exposed to the situations described above (see also accompanying statement from Prof. Baird; the work at Lund will not be funded by this grant, but we will put in place frequent exchange of students between EPFL and Lund University). The models will be initially developed and assessed in physics-based simulations of flying vehicles and of compound eye vision. The most promising models will be validated on small drones with insect-like dynamics, which will be an adaptation of customizable and programmable drones developed in our lab. The visual sensors of these drones will consist of the curved artificial compound eyes developed in a previous project. We expect that this project will result in the next generation of drones capable of visually guided flight through forests, in urban streets, and in formation with other flying robots, which so far has been achieved only by sharing GPS coordinates among drones. We also expect that the ongoing collaboration with insect biologists will shed new light on the behavior and potential mechanisms employed by living insects faced with similar situations.