to filter) the actual sensor measurements (i.e. the state space model) to make small adjustments to (i.e. The Extended Kalman Filter is an algorithm that leverages our knowledge of the physics of motion of the system (i.e. It calculates a weighted average of actual sensor measurements at the current timestep t and predicted sensor measurements at the current timestep t to generate a better current state estimate. y vector).įrom here, the Extended Kalman Filter takes care of the rest. Then we used the current state at time t to infer what the sensor measurements would be at the current timestep (i.e. We started by using the previous estimate of the state (at time t-1) to estimate the current state at time t. I say “predicted” because remember the process we went through above. The y vector represents predicted sensor measurements for the current timestep t. Note: If that equation above doesn’t make sense to you, please check out the observation model tutorial where I derive it from scratch and show an example in Python code. We want to know why we use EKFs.Ĭonsider this two-wheeled differential drive robot car below. What is the Extended Kalman Filter?īefore we dive into the details of how EKFs work, let’s understand what EKFs do on a high level. Otherwise, if you feel confident about state space models and observations models, jump right into this tutorial. These mathematical models are the two main building blocks for EKFs. In order to understand what an EKF is, you should know what a state space model and an observation model are. To get the most out of this tutorial, I recommend you go through these two tutorials first. By “state”, I mean “where is the robot,” “what is its orientation,” etc. Your robot’s sensors are noisy and aren’t 100% accurate (which is always the case).īy running all sensor observations through an EKF, you smooth out noisy sensor measurements and can calculate a better estimate of the state of the robot at each timestep t as the robot moves around in the world.You have a robot with sensors attached to it that enable it to perceive the world.You’ll see them in everything from self-driving cars to drones. EKFs are common in real-world robotics applications.
0 kommentar(er)
