Week 6 (3rd Period)

When trying to simulate both algorithms, the students faced two problems. Firstly, MCL3D proved to be incompatible with the ARM64 architecture, and one of the group members had a computer with this specification. The second problem was limited computing power. That is, when running the algorithms with Gazebo and Rviz simultaneously, the system’s performance became excessively slow.

Initially, it was tried to get over the computational power issue by using Rviz on its own, i.e. without the graphics component (gazebo_gui:=false). However, this approach proved unsuitable, since checking the robot’s current position was a complicated task. So the solution adopted was to use a remote computer from SocRob@Home, whose computing power is significantly higher than the team members’ computers, and which also solved the problem of the computer architecture incompatibility.

Thus, it was time to start the simulation tests, in which various experiments were performed. Firstly, the decision was made to trace the same route manually using both algorithms, in order to guarantee maximum impartiality in the tests. This decision was made because it was impossible for MCL3D to use the Rosbag generated by the Mapping task. However, the students soon realised that this approach did not provide very accurate results, which led them to a new idea. Since the MCL3D didn’t support the execution of rosbags, they decided to record a rosbag running this same algorithm and then apply that rosbag to the HDL. This way, the path travelled by the robot was exactly the same for both algorithms, allowing for more precise measurements.

The gif below shows the HDL algorithm being tested in simulation. The red points represents the sensor reading, and the aim is for this area to overlap the map as much as possible.

Visually, HDL shows adequate and consistent behaviour.

The image below illustrates MCL3D being tested in simulation.

Based on what has been explained previously, it is clear that this algorithm is not ideal for simulation. It appears to be very unstable and shows constant dislocalizations, meaning that the robot doesn’t have a precise idea of its localization.