Closed djp42 closed 5 years ago
We saw an issue elsewhere where the road was misoriented, which could lead to the positions (and thus the distances) being wrong. Alternatively, some of the lidar updates in AutomotiveDrivingModels could have messed with the values.
Old values (displayed using Julia 1.1, so the issue has to be in the extract_demonstrations
):
'0820am-0835am': {
'dleft': (-2.3049260146309027, 21.96607412120762),
'dright': (-2.3457481513705556, 20.482909199293207),
'Turn-rate': (-0.7525520700884558, 0.6810855783650847),
'Acceleration': (-9.169478099589634, 19.6527255780126),
'Jerk': (-56.311584929457936, 170.3583576401261),
'Speed': (4.4628430053130104e-05, 30.40344487830156),
'iTTC': (0.03333333333333333, 1276.2552473386506),
'TTC': (0.0007835423220278857, 30.0)},
After that push to NGSIM, this is resolved (the values are more reasonable):
{'0750am-0805am': {
'Speed': (3.8103724410853456e-06, 29.570213614914064),
'dright': (-1.7140965580048575, 20.555864801174923),
'iTTC': (0.03333333333333333, 1701.604566642411),
'TTC': (0.0005876806042976188, 30.0),
'Turn-rate': (-2.3524487840378727, 1.135634508056773),
'Acceleration': (-13.339110489477832, 39.66206075144301),
'Jerk': (-111.2891908640254, 328.2468851536772),
'dleft': (-1.318445210868404, 21.840951362599814)}}
After running julia 1.1 version of the ngsim extract everything and looking at the expert data in
visualize_emergent.ipynb
, we see completely wrong values fordleft
,dright
, andTTC
, likely the result of some undocumented change in a dependency between julia versions (speculation). Some of the other values also seem a bit off. See below for values: