A UKF-based orientation estimator for the Atlas platform.
详细信息
文摘
The Atlas platform being developed at Carleton University is fully dexterous and unconstrained in the rotational sense. Currently, there are sensors capable of measuring the orientation of the Atlas sphere, however each sensor has certain limitations. Having concomitant orientation measurements from two sensors sets up an opportunity to improve the overall accuracy of the orientation estimate. A method for fusing two measurements can take advantage of this in order to improve the orientation estimate. An algorithm is necessary in order to properly fuse measurements from two sensors, and the algorithm needs to be able to handle rotations characteristic of the Atlas platform. This dissertation presents a novel algorithm for improving estimation of the orientation of the Atlas platform using an adapted unscented Kalman filter (UKF). Two sensors are used due to their complimentary characteristics. The first is the inertial orientation sensor (IOS), which is a common low cost inertial measurement unit (IMU) used for high frequency attitude sensing, that will typically perform poorly over time due to high drift. A second absolute sensor, the Atlas visual orienting sensor (VOS), is a digital camera that operates with a lower frequency, and is used to correct for the inertial sensor's drift. The VOS measures the absolute orientation of the platform, processes the images, and obtains an estimated orientation quaternion, but at a slower frequency of approximately 20 Hz, compared to the IOS which operates at 76 Hz. This thesis outlines the development of a quaternion based indirect UKF for sensor fusion with sensor error estimation and out of sequence measurement (OOSM) handling. The sensor fusion filter obtains an improved estimate given measurements from these two sensors. Due to the unbounded orientation workspace of the platform, representational singularities associated with Euler angles are overcome by utilizing quaternions. IOS stabilized measurements act as direct input to the adapted UKF algorithm and are further improved using statistical information about the gyro scaling factors, gyro misalignments and gyro drift provided by the IMU manufacturer specifications. Within the algorithm, attempt is made for gyro errors to be estimated and corrected using knowledge from the VOS. As well, this algorithm overcomes issues associated to latency between measurements that is the result of measurements arriving out of sequence. Simulation of the filter was conducted accounting for the possibility of OOSMs, measurement noise, and various sensor frequencies. Afterwards, real IOS data was recorded and passed through the UKF to examine the validity of the filter. Results of the simulation and test experiments are detailed and discussed herein. As demonstrated, the developed Atlas UKF improves estimation of the Atlas orientation beyond the capabilities of either sensor alone, and at the same time compensates online for misalignments and drift caused by the IOS.