This project use the package robot_localization to fuse the robot poses estimation. The poses are estimated by two methods: 3D visual reconstruction and ACML (Monte Carlo localization). For sensory fusion to work properly, the center of the ArUco must be positioned at the center of the robot. The figure below shows how the system works, where each element is individually described below:
The Extended Kalman Filter (EKF) is used to fuse the robot pose estimates in this project. Thus, the robot_localization package executes an EKF node that fuse the amcl pose and the 3D visual reconstruction pose and publishes it in the topic odometry/filtred_map. The robot_localization is very well explained in this links:
Amcl is a probabilistic localization system for a robot moving in 2D. It implements the adaptive Monte Carlo localization approach , which uses a particle filter to track the pose of a robot against a known map.
Consume the poses estimated by is-reconstruction and publish them to the ROS is/vo topic.
The Is-reconstruction microsservice estimate the ArUco's pose using the detections from all cameras that are viewing it.
Ros-map-server-microsservice: Provide the environment map for ros and PIS.
ROS-master: Provides naming and registration services to the rest of the nodes in the ROS system.
Is-reconstruction: Estimate the 3D poses of the ArUco Markers.
The deployment file, which is in amcl/etc/k8s, contains the amcl configuration we are currently using. However, if you want to use robot_localization to provide the map --> odom transformation, you must set the tf_broadcast parameter to False (tf_broadcast: False).
The ROS-translation microservice is used for the integration of Ros with the intelligent space. However, this project does not make use of it yet. Thus, the reconstruction_3d package is used to make the is-reconstruction measurements available to ROS.
| Name | ⇒ Input | Output ⇒ | Description |
|---|---|---|---|
| is/vo | 📨 topic: reconstruction.{ArUco_id}.ArUco 💎 schema: Pose |
📨 topic: is/vo 💎 schema: PoseWithCovarianceStamped |
Takes the marker pose and publishes it to ROS with a specific covariance, which is statically defined in the package configuration. |
The deployment file, which is in robot_localization/etc/k8s, contains the robot_localization configuration we are currently using.
Unites all three configurations (amcl, Is-reconstruction e robot_localization) into a single deployment.
ROS is a distributed computing environment. However, it has requirements of the network configuration:
- There must be complete, bi-directional connectivity between all pairs of machines, on all ports.
- Each machine must advertise itself by a name that all other machines can resolve.
This project proposes to execute all the ROS necessary packages in a cluster k8s. This tutorial show a solution to execute ROS nodes in the cluster k8s, but it not works in this project because the robot network does not belong to the cluster network and ROS nodes use TCP/UDP/DDS to communicate with each other on random ports, while k8s wants to expose services on configurable ports and typically over HTTP(S). SO, Ingress controllers don't really play well with ROS.
The initial solution for all this is to declare the ROS packages with the manifest hostnetwork:True. But, any other solutions for this problem are welcome as suggestions :).