ICP-based motion model for AMCL
Hi,
i'm going through the AMCL architecture and i have one question that i can not find an explicit answer to it, i hope that someone can help for it.
In the prediction step of a particle filter, we have the proposal distribution that describe the motion of the robot, the most famous motion model is the odometry-based motion model described in the Probabilistic Robotics famous book (the same model is implemented in the AMCL package). My thinking is the following : When i use ICP scan matching method to compute relative position (odometry), then i provide this results as odometry input in the AMCL package and it will use it as a command (like in the book), so my question is based on that : Is it correct to use odometry resulted from ICP method in the odometry-based motion model (in practice it works very good) or its better to use ICP method directly as a motion model ?
Hoping that my explanation was clear, if not please feel free to ask me more clarification about that.
Thank you