← Back to main

V5 Closed-Loop Simulation (Phase 1: Offline Replay)

각 프레임의 예측 action을 누적해 kinematic trajectory를 생성하고 expert trajectory와 비교. 성공 기준: FPE < 0.5m AND TLD ∈ [0.7, 1.5].

step2
66.7%
success rate
FPE: 0.55m   TLD: 1.03
exp11
0.0%
success rate
FPE: 1.45m   TLD: 1.03
step3
60.0%
success rate
FPE: 0.48m   TLD: 0.97
exp17
11.1%
success rate
FPE: 1.04m   TLD: 1.08
exp18
11.1%
success rate
FPE: 1.04m   TLD: 1.08
exp19
55.6%
success rate
FPE: 0.51m   TLD: 0.95
exp49
96.7%
success rate
FPE: 0.08m   TLD: 1.00
exp50
83.3%
success rate
FPE: 0.24m   TLD: 1.00
exp51
96.7%
success rate
FPE: 0.16m   TLD: 1.00
exp52
93.3%
success rate
FPE: 0.13m   TLD: 1.00
exp54_s2v2
96.7%
success rate
FPE: 0.11m   TLD: 1.01
exp55
96.7%
success rate
FPE: 0.12m   TLD: 1.02
exp54_s2v2_free
0.0%
success rate
FPE: 2.16m   TLD: 1.09
exp53
96.6%
success rate
FPE: 0.13m   TLD: 0.99

Per Path Type

Path Typestep2 successstep2 FPEexp11 successexp11 FPEstep3 successstep3 FPEexp17 successexp17 FPEexp18 successexp18 FPEexp19 successexp19 FPEexp49 successexp49 FPEexp50 successexp50 FPEexp51 successexp51 FPEexp52 successexp52 FPEexp54_s2v2 successexp54_s2v2 FPEexp55 successexp55 FPEexp54_s2v2_free successexp54_s2v2_free FPEexp53 successexp53 FPE
center_straight0% (0/1)0.95m0% (0/1)1.72m100% (4/4)0.29m0% (0/1)1.61m0% (0/1)1.61m100% (1/1)0.33m100% (4/4)0.10m100% (4/4)0.25m100% (4/4)0.22m100% (4/4)0.13m100% (4/4)0.15m100% (4/4)0.18m——100% (4/4)0.09m
center_left100% (1/1)0.36m0% (0/1)1.85m0% (0/3)0.74m0% (0/1)0.70m0% (0/1)0.70m100% (1/1)0.23m100% (3/3)0.08m67% (2/3)0.38m100% (3/3)0.11m100% (3/3)0.04m100% (3/3)0.08m100% (3/3)0.15m——100% (3/3)0.28m
center_right100% (1/1)0.23m0% (0/1)1.14m67% (2/3)0.52m100% (1/1)0.46m100% (1/1)0.46m100% (1/1)0.49m100% (3/3)0.04m67% (2/3)0.42m100% (3/3)0.08m100% (3/3)0.08m100% (3/3)0.08m100% (3/3)0.04m——100% (3/3)0.11m
left_straight100% (1/1)0.23m0% (0/1)1.19m100% (4/4)0.21m0% (0/1)0.82m0% (0/1)0.82m100% (1/1)0.46m100% (4/4)0.00m100% (4/4)0.00m100% (4/4)0.00m100% (4/4)0.05m100% (4/4)0.00m100% (4/4)0.06m——100% (3/3)0.00m
left_left100% (1/1)0.21m0% (0/1)1.80m33% (1/3)0.53m0% (0/1)1.22m0% (0/1)1.22m0% (0/1)0.85m100% (3/3)0.11m100% (3/3)0.19m100% (3/3)0.04m100% (3/3)0.16m100% (3/3)0.16m100% (3/3)0.00m——100% (3/3)0.08m
left_right100% (1/1)0.38m0% (0/1)1.18m67% (2/3)0.50m0% (0/1)0.98m0% (0/1)0.98m0% (0/1)0.63m100% (3/3)0.12m100% (3/3)0.08m100% (3/3)0.47m100% (3/3)0.20m100% (3/3)0.12m100% (3/3)0.04m——100% (3/3)0.24m
right_straight100% (1/1)0.34m0% (0/1)1.26m100% (4/4)0.25m0% (0/1)1.12m0% (0/1)1.12m100% (1/1)0.23m100% (4/4)0.00m100% (4/4)0.00m100% (4/4)0.00m100% (4/4)0.00m100% (4/4)0.03m100% (4/4)0.09m——100% (4/4)0.00m
right_left0% (0/1)0.52m0% (0/1)1.91m0% (0/3)0.85m0% (0/1)1.17m0% (0/1)1.17m0% (0/1)0.73m100% (3/3)0.00m100% (3/3)0.32m100% (3/3)0.23m100% (3/3)0.11m100% (3/3)0.00m100% (3/3)0.12m——100% (3/3)0.04m
right_right0% (0/1)1.77m0% (0/1)1.03m0% (0/3)0.79m0% (0/1)1.31m0% (0/1)1.31m0% (0/1)0.60m67% (2/3)0.32m0% (0/3)0.69m67% (2/3)0.39m33% (1/3)0.42m67% (2/3)0.39m67% (2/3)0.42m——67% (2/3)0.39m
설정
Phase 1: offline replay — 원본 H5 이미지로 예측, 예측 action을 kinematic model에 적분.
속도 매핑: lx/ly=1.15 m/s, az=±0.25 rad/s (데이터 실측). dt=0.1s.
성공 기준: FPE < 0.5m AND TLD ∈ [0.7, 1.5].