I have decided in my spare time (and potentially a co-op in the near future) to focus some of my energy on building an autonomous quadcopter. I have a passion for building things that are smart and autonomous and I figure a large scale robotics project is the next step. Since January I have been researching and flushing out the details, requirements, and challenges that I'll have to overcome to implement this. The planning stage has been a great opportunity to put to practice some of the skills I'm learning on my classes such as Project Management and problem analysis.
The problem is a complex one. I had the idea when I was mountain biking last summer, and thinking about how complex sports videography is getting. I was thinking about cable-cam systems and how much effort is put into the deployment of such as system, especially in an evironment such as a forest. I thouight it would be great to have a camera that flies around you, dodging obstacles, and keeping the rider in sight, with no interaction from the user. This, along with a nagging to build a large scale robot project, formed the core of my idea.
I will develop a flying camera system designed for the extreme sports market to autonomously film and follow a target subject. I began researching platforms and existing implementations, and found that I was not the first person to mount a camera on a helicopter (duuhhh), not the first person to use a multi-rotor camera system in an extreme sports environment, nor the first person to build an autonomous multi-rotor copter. But, I found no documented projects that addressed all the problems I hoped to solve, nor had I built any of them (lol).
So I decided the project was a go, and began planning my solution. I discovered man y prefabbed kits, some with autonomous cability built in, some without. I eventually settled on the kits provided by DIYDrones. These kits implement the ArduCopter/ArduMEGA system. This system has autonomous capability out of the box, using open source software. Open source will allow me to easily modify (and contribute), as well as find support for issues I'll encounter. When the idea was first imagined, I considered the idea of using a Kinect module to scan and map obstacles, providing a rich 3D read out of the immediate environment. Unfortunately it did not have the speed, distance, or accuracy that I desired, as well as the fact that it was quite heavy, and mounting it along with the camera would be potentially complicated. I settled on some small untrasonic distance sensors, providing a wide field of view and accuracy to 1 inch. With an array of these sensors, the copter could have an effective rough idea of nearby obstacles and their distance. Some testing will be required to test the sensors sensitivity to sudo-objects such as tree branches, but I believe they will be effective.
Check the blog for further updates.