Mobile robot localization is the problem of determining the pose of a robot relative to a given map of the environment. In other words, with knowledge of the map of the environment, localization places the robot in the map, with the help of sensor data, motion models and control inputs. Localization is naturally very important for mobile robot navigation. The Gaussian filtering algorithms are one of the categories of algorithms used for mobile robot localization. For this project, I have implemented the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF) algorithms for mobile robot localization and I compared their performance. Both of these algorithms are Gaussian filtering algorithms, however, ExKF is based on the idea of linearizing whereas UKF is an integration-based Gaussian filtering algorithm.
I have implemented local, passive localization in a static environment for a single robot. Local localization is the case of localization when the initial pose of the robot is known. Passive localization means that the control law of the robot’s motion is not designed specifically for localization; the motion might be random. I have implemented these algorithms on the UTIAS Multi-Robot Cooperative Localization and Mapping Dataset. This dataset contains the following relevant data for 5 robots:
- Odometry: The odometry data has the forward and angular velocity of the mobile robot.
- Measurement data (range and bearing): This contains estimates of distances and angles at which the landmarks were detected by the robot. Accurate groundtruth data: The groundtruth data is the true pose of the robot as it moves around in the environment.
- Position and identity of landmarks: The true positions of all the landmarks is provided. Further, each landmark is linked with a unique barcode which the robot detects while obtaining the measurement data. Thus, the robot can recognise the landmark being observed and place itself on the map accordingly.
This dataset includes data for 5 robots. Since I am only performing localization for one robot, all measurement data corresponding to other robots will be ignored. This can be done because like the landmarks, all robots also have a unique barcode assigne to it, hence the robot can detect whether it is detecting a landmark or another robot.
For further details, please refer to the project report.