If one is interested in implementing a path planning algorithm that is grid-based, one needs to consider the fact that your grid points will never represent the true state of the robot.

How is this dealt with?

Suppose we're doing path planning using a grid-based search on the side of the control for a desired grid position as an output state.

How would you handle the discrepancy between your actual starting position and your discretized starting position?

I understand that normally you may use an MPC instead, which continually recalculates an optimal path using some type of nonlinear solver, but suppose we don't do this - suppose we restrict ourselves to only a grid search and suppose at after every action the state of the robot has to be considered as living in a particular grid point.

  • $\begingroup$ So, will your robot use some kind of approximate odometry information that can be subject to drift and you will be performing path planning as the robot navigates the environment? $\endgroup$
    – nbro
    Commented Mar 11, 2020 at 2:59
  • $\begingroup$ Not just drift, in other words, suppose your robot only knows what cell it is in but not its precise location within the cell. I guess the issue is "robust planning. After each action you can always assign the robot to it's closest cell, thus you could repeatedly call the planner after every action to address this problem but it will be VERY computationally expensive. So how does one do robust path planning? $\endgroup$ Commented Mar 11, 2020 at 6:58


You must log in to answer this question.