Location:Training / Testing and Measurement / Learn about sensor fusion and tracking / Fusion of GPS and IMU to estimate pose

This video provides an overview of what sensor fusion is and its application in autonomous system design. The video also covers scenarios that illustrate the different ways to implement sensor fusion.
Sensor fusion is a critical part of confirming localization, location, and detection and object tracking. We will show that sensor fusion is more than just a Kalman filter. It is a set of algorithms that can fuse data from multiple sources to better estimate the state of a system. The four main benefits of sensor fusion are improved measurement quality, reliability and coverage, and the ability to estimate states that are not directly measured. The fact that sensor fusion has such broad appeal across completely different types of autonomous systems makes it an interesting and rewarding topic to study.

This video explains how to estimate the orientation of an object using a magnetometer, accelerometer, and gyroscope. The aim is to show how these sensors contribute to the solution and explain some considerations.
We'll cover what orientation (or attitude) is and how to determine it using accelerometers and magnetometers. We will also discuss calibrating the magnetometer for hard and soft iron sources, as well as methods for dealing with acceleration damage.
We will also demonstrate a simple dead reckoning solution that uses a gyroscope alone. Finally, we will introduce the concept of a hybrid three-sensor solution.

This video continues our discussion of using sensor fusion for point localization and relative position localization by showing how to estimate the orientation and position of an object using GPS and IMU. We will introduce the structure of the algorithm and show you how GPS and IMU work together in the final solution, giving you a more intuitive understanding of the problem.

This video shows how to track a single object by estimating the state from multiple model filters that interact. We build some intuition about the IMM filter and show that it is a better tracking algorithm than the single-model Kalman filter.
The part that makes tracking harder is that we usually have to do it with less information. We will introduce how the IMM filter can compensate for the lack of information and show some simulation results.

This video covers two common issues that arise when tracking multiple objects: data association and tracking maintenance. We introduce several approaches to solving these problems and provide a general approach to all multi-object tracking problems.
We introduce data association algorithms such as global nearest neighbor (GNN) and joint probabilistic data association (JPDA), and introduce criteria for deleting and creating tracks. We will also talk about gated filtering mechanisms to avoid wasting computational resources. At the end of the video we show an example of the GNN and JPDA algorithms operating on two objects that are in close proximity.

A deeper understanding of trajectory-level fusion (or trajectory fusion), the types of tracking situations that require it, and some of the challenges associated with it.
You'll see two different tracking architectures - rail-to-rail fusion and center-level tracking - and understand the benefits of choosing one architecture over the other.