Other Rewards

No failure penalty

NGSIM

Successful

success, right behind his back 28.07.2020_17:06:18_11_ckpt=06000_scenario=NGSIM_LANE_CHANGE_I80_VALIDATION_steps=34_total-reward=1.9.avi

Awaiting impatiently for the opportunity

(again, reference driver did it slower, smoother transition)

success, but slowing down 28.07.2020_17:06:18_16_ckpt=06000_scenario=NGSIM_LANE_CHANGE_US101_VALIDATION_steps=28_total-reward=1.9.avi

Correct, but slowing down after

(reference driver hasn't even started his maneuver)

Failed

failed, tried 3 times, almost did it 28.07.2020_17:06:18_17_ckpt=06000_scenario=NGSIM_LANE_CHANGE_US101_VALIDATION_steps=36_total-reward=-0.19999999999999996.avi

Unique example: 2 attempts
3rd almost successful

(steering too wobbly)

failed, cut-in situation 28.07.2020_17:06:18_6_ckpt=06000_scenario=NGSIM_LANE_CHANGE_US101_VALIDATION_steps=15_total-reward=-0.30000000000000004.avi

Cut-in, more than one lane change

(actually one and a half, agent was spawned exactly between two lanes, seemingly learned that maneuver must not be too short in order to get a reward)

Successful

success, aggressive acceleration, slipped 28.07.2020_15:12:55_0_ckpt=03600_scenario=OPENDD_RDB6_DENSE_steps=150_total-reward=2.0.avi

Slipped while breaking to keep distance

(was able to wait until the car in front of him moves off)

success, snappier than reference driver 28.07.2020_15:12:55_1_ckpt=03600_scenario=OPENDD_RDB3_DENSE_steps=98_total-reward=2.0.avi

Normal entering, snappier than reference driver

(had to hit the brakes a bit at roundabout)

Failed

failed, traffic jam, impatient for reward 28.07.2020_15:12:55_9_ckpt=03600_scenario=OPENDD_RDB7_DENSE_steps=21_total-reward=-1.avi

Impatient in traffic jam

(stuck for too long, decided to move off, afraid of timeout)

failed, going off track 28.07.2020_15:12:55_5_ckpt=03600_scenario=OPENDD_RDB1_DENSE_steps=57_total-reward=-0.6.avi

Moved too far after recovering from small slip

(rare example considering trained agent)