• UFRJ Nautilus

Gazebo: the Power of Simulations

Updated: May 7


In many automation applications, specially in the case of autonomous vehicles,

like the AUV we develop, doing tests with the actual equipment is not always a viable option.

These endeavours usually take very long and often require a big team and a place with

adequate conditions, which might mean very high operational costs.

In spite of said problems, virtual simulation tools have become very capable and

useful when developing engineering projects nowadays. With that in mind, we’ll briefly

explore one of these tools: Gazebo.

Gazebo is a simulation program mostly employed in the robotics sector and its main

feature is being capable of calculating physics and graphics in real time while maintaining its

simplicity. Furthermore, its compatibility is one of its main selling points, specially in the

case of systems using ROS, which is known to be widely integrated with the Gazebo

ecosystem. This makes it so the environment as a whole is ideal for debugging algorithms of

many kinds.


Simulation in the Gazebo


In order to add to the realism of the simulations provided by Gazebo, it’s possible to

enlarge its feature set through the use of plugins. These plugins are often, but not restricted

to, used to mimic the behaviour of specific sensor. The simulated sensors then can be fine-

tuned so that their output is indistinguishable from the real world, and therefore enabling a

single to source code to power both the real machine and the simulated one, eliminating the

need of tweaking their inner-workings.

Regarding its use within UFRJ Nautilus, Gazebo is often used to try improvements to

the AUV’s state machine without the need to take the vehicle to an actual pool. This

enables software features to be worked on in an iterative manner, trying new algorithms in

a dynamic workflow.


Simulation of our AUV in the Gazebo

Beyond this use, Gazebo is also employed in many machine learning applications,

being particularly useful in reinforcement learning tasks, that rely heavily on simulated

environments. This happens because this type of use often needs to assign scores,

incentives or punishes to an agent in a given task, which is made easier by Gazebo’s high

flexibility. This aspect, alongside its compatibility, make this tools potential virtually limitless

in this field.


Simulation of a Mechanical Arm in the Gazebo