Robotics Stack Exchange is a question and answer site for professional robotic engineers, hobbyists, researchers and students. It only takes a minute to sign up.
Sign up to join this community
I have studied Claus Brenner's lectures on how to implement the FastSLAM 1.0 algorithm, where each particle maintains the robot pose, and maintains EKF's of the landmarks.
However, I'd like to implement FastSlam 2.0. Which I understand uses particle filters completely. Is this the same as FS 1.0, but instead of each particle maintaining an EKF of the landmark, each particle maintains yet another array of particle filters for landmarks?
I suggest you to give a look to Sebastian Thrun's work here. In fastSLAM (both 1.0 and 2.0) each particle maintains an array which contains the states of the landmarks as well as the robot's states. The main novelty of fastSLAM 2.0 consists in the update of the robot's pose which does not only keep into account of the odometry but also of the most recent measurement.
FastSLAM 2.0 separates the estimation of the state of the vehicle from the state of the map, which in this case, is the set of landmarks one is using to self-localize.
If one were to use a grid based variant of the fast slam one would not need to estimate the state of each landmark via an EKF however, each particle should keep a its own estimate of the map. This is followed by a resampling step
