This module contains implementations of prior works used for comparison against iMESA. Below are descriptions of each supported prior work.
baselines/- Contains implementation of the Centralized and Independent baselinesddfsam2/- Contains implementation of DDF-SAM2
This baseline represents that which could be achieved if all robots communicate their measurements back to a central server. Under the communication model assumed in the paper above, such an architecture is impossible. However, a centralized solver provides a level of accuracy that we strive to match with distributed methods. Thus it is an excellent baseline for comparison. Note: to implement the centralized solver within the interface of IncrementalSAMAgent we had to get fancy with static variables. Modify the this implementation with care.
This baseline represents that which could be achieved if robots never communicate, and instead each independently optimize their local factor-graph. This baseline should represent an upper-bound on the performance of a distributed method.
This prior work was originally presented in the paper "DDF-SAM 2.0: Consistent distributed smoothing and mapping" by Cunningham et al. in 2012. In this approach robots share potentials being careful to avoid double counting of information. A significant downside to DDF-SAM2 compared to iMESA is that DDF-SAM2 does not enforce robots converge to a single solution, and the necessity to compute marginals negatively affects its runtime efficiency. Note: we use the naive bayes approximation as suggested by Cunningham et al.