Sensor Fusion and Task Localization System

[stepheffects]

The sensor fusion system should provide a state estimation for the decision making system to allow task localization. The system should take as input the output from each sensor and the vision system and return the inferred state the sub is in for the competition. The system will use a dynamic Bayesian network operating at at an time step determined by testing. The system will use an appropriate inference algorithm such as Gibbs sampling or Monte Carlo inference methods and use appropriate learning techniques to adapt the sensor errors to their true values based on test data.

[Azimath]

Consider including some provision to set the time factor based on a faster sensor and use extrapolated data from the slow ones in the event our slowest sensor is too slow. Being able to test with the slowest sensor as the time base vs a fast sensor with extrapolation on the slow data is probably important.

[Azimath]

The way ROS likes to handle fast vs slow sensors is to have a "local" frame that uses fast sensors that are only accurate in the short term and a "global/correction" frame that moves the origin of the local frame around to correct for accumulated error relative to the world using data from slow but stable sensors.