Simulation environment

RT-SLAM has a high level simulation environment.

It allows to remove all the hardware sensor and raw data processing steps in order to control errors, noises, and ground truth. This is important for debugging and evaluation.

This simulation environment is basic and simple, and is complementary to a real versatile simulator such as MORSE .

The simulation environment is composed of two parts:

• the simulator itself, that contains a description of the scenario and the environment (robots, sensors, landmarks, robots trajectories).
• implementation of abstract classes for simulation : hardware classes that use the simulator to provide outputs, and high level simulated data and raw processors.

The simulator

The simulator part is composed of the adhoc simulator itself, that provides an interface from outside (in file simulator.hpp):

• simu::AdhocSimulator

And some internal objects that are used for the simulation (in file simulatorObjects.hpp):

• simu::Robot a simulated robot, which contains a description of its trajectory during the experiment.
• simu::Sensor a simulated sensor, which contains observation models for the different types of landmarks it is supposed to observe.
• simu::Landmark a simulated landmark

The implementations of abstract classes

Because we simulate hardware, a simulation-targeted implementation of all classes that are related with hardware must be provided:

• simu::HardwareSensorAdhocSimulator (in file hardwareSensorAdhocSimulator.hpp)
• simu::HardwareEstimatorInertialAdhocSimulator (in file hardwareEstimatorInertialAdhocSimulator::hpp)

And because we also simulate high level features and are working with semantic sensors (ie that direcly provides observations of points, lines...), we need to provide simulation-targeted implementations of all data structures related to raw data (in file simuData.hpp):

• simu::RawSimu
• simu::AppearanceSimu
• simu::FeatureSimu
• simu::DescriptorSimu

As well as the raw processors that work with this new type of raw data (in file simuRawProcessors.hpp):

• simu::DetectorSimu
• simu::MatcherSimu

Fortunately as we are working with high level features, all of these classes are extremely simple.

Management of errors and noises

Errors and noises can be injected locally:

• raw data processing (detection and matching): set noise in rawProcessors
• extrinsic transformations: set a slightly different transformation when building simulation objects than initializing slam objects.
• hardware estimators: set noise, biases and gains in simulation hardware simulators.