A Biomimetic Approach to an Autonomous Unmanned Air Vehicle
(or how to create an artificial insect with an Ar.Drone)
Create two sets of modules, one for the input (vision) and one for the output (movement), which would permit the use of the quadrotor with a neuronal simulator (iqr).
Apply existing studies of the visual system of the insects in order to propose an approach that learns from nature.
Validate the models that derive by these studies in a real world situation.
Having to create a fully autonomous flying robot is a challenging task. Proposing a biomimetic approach makes it even harder.
For my master thesis in the Cognitive Systems and Interactive Media MSc at Universitat Pompeu Fabra (UPF) of Barcelona, i had the following tasks:
After the completion of my thesis i worked in SPECS group of UPF in order to deliver a complete set of modules and a more complete study.
This has led to the extended abstract of the study, which has been published in the proceedings of the 1st Living Machines conference in Barcelona (2012). You can find the study here
- The arena
- Thesis Method
- In flight
So far, iqr can communicate with Ar.Drone by using the following modules:
Bottom camera, represented by a 44x36 neuronal group (RGB, HSV)
Frontal camera, represented by a 64x46 neuronal group (RGB, HSV)
Simple control module, controlling the 4 axis (pitch, roll, yaw, altitude)
NavData control module, adding up the navigational information from the sensors of the robot, like altitude, axis angles, battery etc.
The models used, manage to retrieve information about the speed, direction and looming stimuli from the environment, by using a sole sensor; the camera.
This approach is able to get basic navigational information and behavior without using external sensors and with a proper behavioral layer to achieve task autonomy.
The video below demonstrates the functionality of the modules, where the drone follows the blue ball (pitch) and moves according with the red (yaw, gaz).
The second part demonstrates the experiments made by using the models of the visual system of the insects (fly, locust).