The basic ideas behind the operation of the MonteCarlo localization algorithm are:
- Prediction phase: The objective is to use the information about previous location estimations to obtain new estimations about the current positions.
- Update phase: based on the received observations, the weights of the nodes are updated.
- Normalization phase: the new weights are normalized to ones and zeros, so that the posterior distribution can be simulated.
The prediction and update phases contain recursive calculations that depend on data obtained from the previous calculations. After the normalization phase, the weak weights are discarded since we only want to concentrate on trajectories with the larger weights. If too many samples are discarded, and the current number of samples is below certain threshold (typically 50) then a re-sampling is made to keep enough valid samples.
The concept of resolution limit is introduced, and it refers to the probability that a node can move a distance d without changing connectivity. This is an important parameter for this technique.
The implementation of security in this algorithm is more feasible than with other techniques, since it supports bidirectional verification, key establishment and there are continued location estimates. When nodes and seeds move, rogue nodes can cause only limited damage.
The algorithm was evaluated extensively and compared with the performance of other techniques as Centroid and Amorphous, particularly, the accuracy is the key metric in all experiments. MCL outperforms the other techniques in accuracy, when seed and/or node density increases, when the range presents irregularities, but it is greatly affected with group motion, so in the later case, motion control is required.
The main result is surprising and counterintuitive, mobility in this algorithm, can improve accuracy and reduce the costs of Localization, even with severe memory limits, low seed density and irregular node transmissions. Future work is required regarding security and types of motion.
The proposed MCL is a commonly known localization algorithm.MCL periodically updates its samples, which are a node's probabilistic distribution of locations. I think MCL is based on inappropriate hypotheses like MCL requires knowledge of the radio range and max speed of nodes. In reality, the radio transmission ranges are always changing and max velocity of each node cannot be defined in the real world since every node may have different speed limits.This algorithm is well known in mobile sensor networks.
ReplyDeleteI do not agree with Haritha, since I feel the paper does not assume the knowledge of exact speeds of each node in the network. It only bases its calculation on the maximum speed that any node CAN gain in the worst case, in a time frame, which is always available. However, I do think that the paper does not account for the low computational capabilities of sensor nodes.
ReplyDeleteAccording to the results included in the evaluation section, this MCL technique outperforms other Centroid and Amorphous techniques in various aspects and conditions, so I really though of this paper as one very popular and accepted tecnhique. However as we have gone through many papers already, I am starting to realize that the evaluation sections are always claiming great advantages over the other techniques, which might not always be the case.
ReplyDelete