|
Cooperative Localisation Filters 0.0.1
|
Repository containing the implementation of Bayesian filters used for multi-robot cooperative positioning, including:
This repository contains the Cooperative Positioning Data Handler, as a submodule. Therefore, to clone the repository, the following commands are required:
To run a one of the aforementioned filters, the user can select the filter by setting the Filter environment variable before executable:
<Filter Name> can be either: EKF, IEKF, or INFO.
To select a dataset, the user can set the Dataset environment variable before execution:
Dataset can be any value in the range [1,9]. Note that dataset 9 has incorrect groundtruth values, and therefore it is not recomended to use it.
The measurement update can be performed either in a coupled or decoupled manner. The coupled approach creates an augmented state vector and covariance matrix containing the state and covariance of both the ego vehicle and the agent observed respectively. The decoupled approach adds the covariance of the agent observed to the ego robots measurement noise. Since the state of the agent in marginalized out in the coupled approach, both approaches result in identical inference. However, the decoupled approach is less computationally expensive.
The decoupled is the default, however, the user can compile the project using the coupled approach during linking by adding the COUPLED:
The measurement update can be performed using the Huber loss function for both the EKF and IEKF, which results in improved robustness against outliers. To use the Huber loss function, the user must compile the project using the ROBUST flag:
1.9.8