Localization/particle filter
From HotDec
The main files are:
Due to difficulties in the implementation, changes were made to the dictated theoretical outline of the implementation.
- Since the hovercraft floats on a cushion of air, gravity effects the force applied to each thruster.
- As a result, if we include the forces in the motion model update we have particles that travel away from the force of gravity.
- This makes our particle cloud constantly drift away from the true pose.
- Attempts were made to offset the influence of gravity in the calculations, but the effect of gravity is different throughout the room.
- We could not get consistent results with this method.
- We therefore do not include the forces in the motion model update.
- We realize that this violates the theoretical objective of closed-loop control and estimation.
- However, we have designed the measurement model to compensate for this violation by making it more robust and exact.
- Also, we increase the amount of system or motion noise in our motion model to compensate for the effect of the forces.
- Practically speaking, this is a Hovercraft that can't move that far in 50,100 or even 250 ms intervals which is the rate of the filters.
- Hence, using sufficient random noise gives a good particle cloud dispersion and compensates for the neglected forces.
- In practice, this method produces good results in pose estimation, and satisfactory results in Hovercraft control.
