3D Planning for Quadrotors
Typical quadcopter projects use off board systems of high speed cameras to track quadcopters. These are expensive and limits the research in a designated testing area - normally a netted cage. In 2014, the LAIR started a project with the end goal of localize multiple quadcopters with on board systems. This would allow quadcopters to be used as a testbench outside of their cage and increase the range of research problems that could be tested with a quadcopter.
However, to assess the performance of any off board system, there needs to be a system that serves as a basis of comparison. During the summer of 2014, students Sherman Lam ('16) and Jingbin Yang ('15) researched the development of an inexpensive off-board tracking system using the Xbox Kinect for 3D position information. Following the theme of an inexpensive testing system, Bitcraze's Crazyflie quadcopter was chosen as the testing platform.
Sherman optimized a four degree of freedom color tracking system based on Professor Dodd's research in the CS department. He wrote the framework for a 1 degree of freedom (and later a 4 degree of freedom) controller for hover control of the quadcopter. Each of these were based on the Proportional-Integral-Derivative (PID) controller. Both systems where tested on simulated models of the Crazyflie to verify the controllers' stabilities. The tuned four degree of freedom controller was able to sustain the quadcopter at a three minute hover.
Jingbin developed an alternate means of tracking based on the Kinect's depth image. She also explored using a particle filter and a Kalman filter to improve the position estimation of the tracking system. Various resampling and position propagation techniques were explored in conjunction with these filters to find one that would provide the fastest, yet accurate, position estimation.
Here are some of the preliminary results of the 2014 summer research. Hover control was realized with a 4 degree of freedom controller using color tracking and a lowpass filter for position estimation.
While the 4 degree of freedom controller is capable of sustaining the quadcopter at a stable hover, the accuracy of the control is believed to be limited by the control loop update rate (30Hz). This limit is set by the Kinect's data update rate. In addition, using complex position estimation techniques such as a particle filter will decrease the control loop rate, add latency to the position estimation, and decrease the accuracy of the controller.