The Spring 2006 Robotic Capstone course attempted to make a robotic blimp autonomous. Starting with a custom built gondola attached to a 5.5 ft blimp, the class divided into four parts. Each small group would construct a seperate module that when assembled together would create a fully demoable blimp. The four groups were: Control, Localization, Simulation and Visualization.
Control
Control divided into two group: Low-level control and path-planning. Path-planning found an obstacle free path through the atrium to a goal location. Low-level control took tracking locations and goal locations as input and then piloted the blimp to the goal.
Low-level control was found to be surprisingly difficult. Inaccurate input headings presented the most significant challenge, but were eventually overcome to produce a system that greatly out-performs a human operator given the same information.
A write-up of the experiences and the system produced is available here.
Localization
Localization created a system that could track the blimp through the Atrium in the Paul G. Allen building. Using three wall-mounted cameras and a particle filter, the blimp could be accurately localized in 3-space. A red cone attached to the front of the blimp was used to determine the blimp's heading, but had poor accuracy.
Simulation
Using the motion-capture laboratory and Gaussian Processes this group constructed a model for the blimps movement. Initial models analyzed the blimp while freely moving and later models incorporated engine accelerations.
Visualization
This group used an open-source 3D graphics engine (OGRE) to create visualization of the blimp flying through the atrium.