partitioning cameras, landmarks go to BOTH parts (middle of edge = lm) edges are 2 residuals.
- 1/2 r_i for DRS?
- best fill up as what i had before for PALM
- double check: f(x1,..,xi,..xn) -> wrt xi.
- drs sum_fi() is different significantly fi(x1,x2) -> deriv wrt to both. 1/2 fi(x1,x2) -> still to both .. 1/2(fi(const, x2) + fi(x1,const)) gain to split landmark could be to get FULL update locally per lm, problem need to add cams as well