-
Notifications
You must be signed in to change notification settings - Fork 0
/
checklist
39 lines (25 loc) · 1.2 KB
/
checklist
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
-Solve Leaf tests : done! (found D,E,a,c for Schur)
-InnerProducts : computes inner products solely on the blocks corresponding to the states and controls at current and next time steps
- We calculate B_tag and b_tag
how to check this if not within np.dot/dgemm?
-FctorizeBbar (we just have linalg.cho_factor), just computes cholesky fact of B tag and b_tag(this is in soln loop)
-SolveBbar (just have cho_solve), solve cholesky fot y
-UpdateSchur (maybe need to check updateShur?), solve x and z
OCT 10:::
-- check matrix structure for small example (N=2/N=4/N=1?)
--call np.linalg on traditional KKT, on BCHOl and on BCHOL matrix
//Adding BCHOL inside solveKKTSystem - adding 0 to g
Oct 14::
--building the matrix for example and getting the same answer
Oct 16::
-- NP.Linalg == BCHOL!
now you have a ground truth building form Q,R and etc
Next things to do: integrate your Bchol into pendulum
check your soln for other timesteps (4,8,16,32...)
comeback to Cuda and clean/optimize code
First things first :
- add it to pendulum
Oct 21:
Added BCHOl and LQRbuild to TrajopMPCReference - throws an error on level 1 (is_choleskysafe)
- fix buildBCHOL from blocks or delete it
- test BCHOL for different examples