The measurement model should multiply the probabilities assign to each LiDAR end point, but in practice this is numerical instable and results in underflows. The current workaround is to sum the probabilities, but doing so doesn't really make sense theoretically. An investigation needs to be done on how to implement the algorithm in a numerically stable way.
The measurement model should multiply the probabilities assign to each LiDAR end point, but in practice this is numerical instable and results in underflows. The current workaround is to sum the probabilities, but doing so doesn't really make sense theoretically. An investigation needs to be done on how to implement the algorithm in a numerically stable way.