Unable to run properly on TUM-VI dataset

[zichao-shu]

Hello,

I've encountered an issue while using the exp_tum.launch to run the TUM-VI dataset, such as the room scene. The nonlinear optimization does not seem to run correctly. (When I use exp_eth.launch to run the EuRoC dataset, it runs without any problems.) The log information related to this issue is as follows:

[ INFO] [1705805883.917849622]: [VICON-GRAPH]: begin optimization
Warning:  Levenberg-Marquardt giving up because cannot decrease error with maximum lambda
converged
errorThreshold: 2.70976e+10 <? 0
absoluteDecrease: 0 <? 1e-30
relativeDecrease: 0 <? 1e-30
iterations: 0 >? 30
[ INFO] [1705805887.771673091]: [VICON-GRAPH]: done optimization (0 iterations)!
[ INFO] [1705805887.779034128]: [TIME]: 1.0088 to build
[ INFO] [1705805887.779087709]: [TIME]: 4.0854 to optimize
[ INFO] [1705805887.779097869]: [TIME]: 5.0941 total (loop 0)
[ INFO] [1705805887.779122074]: Publishing: /vicon2gt/optimized
[ INFO] [1705805887.779137393]: Publishing: /vicon2gt/vicon

The L-M algorithm ends after 0 iteration, and the generated trajectory is also an incorrect result.

I would greatly appreciate it if you could provide some insight into why this might be happening. Could it be an error in my usage of the tool?

[goldbattle]

I would check your GTSAM version, I know the newer version have issues that I have not had time to debug.
The readme should have the exact version you need to use in it.

[zichao-shu]

Thank you very much for your reply, but I have used the appropriate GTSAM version 4.2a5 with reference to the readme. Also, I'm able to run exp_eth.launch properly for the transformation of the mocap trajectories for the EuRoC dataset, but I have problems with exp_tum.launch for the TUM-VI dataset. Specifically, the nonlinear optimization seems to just end without performing the iteration, and the resulting extrinsics and trajectory are not as expected. Do I need to change some default parameter values in exp_tum.launch (I only specified the path to the rosbag and didn't change any other settings)? Or is there other solutions? Looking forward to your reply.

Also, a more detailed run log is as follows, where you can see that the resulting parameters and time offset are both 0, which I don't think should make sense.

[ INFO] [1705886022.830316090]: [VICON-GRAPH]: graph factors - 28003
[ INFO] [1705886022.833418410]: [VICON-GRAPH]: graph nodes - 14006
[ INFO] [1705886022.943839918]: [VICON-GRAPH]: begin optimization
Warning:  Levenberg-Marquardt giving up because cannot decrease error with maximum lambda
converged
errorThreshold: 1.28089e+10 <? 0
absoluteDecrease: 0 <? 1e-30
relativeDecrease: 0 <? 1e-30
iterations: 0 >? 30
[ INFO] [1705886024.838327784]: [VICON-GRAPH]: done optimization (0 iterations)!
[ INFO] [1705886024.841918418]: [TIME]: 0.5944 to build
[ INFO] [1705886024.841962641]: [TIME]: 2.0081 to optimize
[ INFO] [1705886024.841985144]: [TIME]: 2.6025 total (loop 0)
[ INFO] [1705886024.842009129]: Publishing: /vicon2gt/optimized
[ INFO] [1705886024.842016443]: Publishing: /vicon2gt/vicon

======================================
state_0: 
m_time:[1520530308.69968]'
q:[0.00295727509592, 0.0110676132374, -0.0339767977395, 0.99935696309]'
bg:[0, 0, 0]'
v:[-0.00253357316322, -0.0167442626002, 0.00192056851242]'
ba:[0, 0, 0]'
p:[0.836550118376, -0.230885431553, 1.24700020276]'

state_N: 
m_time:[1520530448.70955]'
q:[0.052349482075, -0.0191554566416, -0.998443389658, 0.00184332739918]'
bg:[0, 0, 0]'
v:[0.0244057842063, -0.0644686179124, -0.0284375890697]'
ba:[0, 0, 0]'
p:[0.156843225345, -0.0604932033898, 1.10624209207]'

R_BtoI: 
1 0 0
0 1 0
0 0 1

p_BinI: 
0
0
0

R_GtoV: 
1 0 0
0 1 0
0 0 1

t_off_vicon_to_imu: 
0

======================================