Skip to content

Releases: team-fusionx/CarND-Capstone

Revised release for submission FX-v3.2

28 Aug 06:36
c1431d9
Compare
Choose a tag to compare

Revised release for submission for Udacity testing. Includes the following improvements:

  • Increase waypoint follower pure pursuit min lookahead distance to 6m to reduce initial steering correction amount
  • Pass images into a deque to keep up to 3 images ready for processing by classifier to increase number of light state detections
  • Extract waypoints from Carla site test data to confirm steering response is still enough to follow the turn around the course
  • Improve RViz visualization for site test including LIDAR point cloud and rescaled markers

Revised release for submission FX-v3.1

15 Aug 16:29
e97febb
Compare
Choose a tag to compare

Revised release for submission for Udacity testing. Includes the following improvements:

  • Added fallback light classification using magnified image clips in case of uncertain initial detection to improve accuracy from farther away
  • Increase speed threshold to transition to creep speed and improve stopping distance margin when light is detected at low speed
  • Increase steering filtering constant to slow down and reduce initial steering correction to get on waypoints at start of test

Revised release for submission FX-v3.0

09 Aug 19:21
898ca30
Compare
Choose a tag to compare

Revised release for submission for Udacity testing. Includes the following improvements:

  • Split traffic light classifier model for sim vs site test by launch file parameters
  • Trained new site test classifier using previous site test ROS bag data that splits images into 5 clipped regions to zoom in on light and does a batch inference to judge the light color
  • Increased tl_detector loop rate to 20 Hz

Revised release for submission FX-v2.4

19 Jul 23:26
Compare
Choose a tag to compare

Revised release for submission for Udacity testing. Includes the following improvements:

  • Check for light waypoint previously stopped at to prevent double stopping at very short light cycles
  • Move throttle/brake PID parameters to launch files instead of dynamic reconfigure to allow separating between simulation and site test
  • Adjust throttle/brake P gains based on Carla's driving response in bag data

Revised release for submission FX-v2.3

10 Jul 02:54
d7520d3
Compare
Choose a tag to compare

Revised release for submission for Udacity testing. Includes the following improvements:

  • Increased traffic light classifier confidence threshold to reduce false RED detections.
  • Add RED->UNKNOWN transition to be allowed if car is already stopped for the red light and light state becomes unknown (possible green light).

Revised release for submission FX-v2.2

22 Jun 19:50
a5afd37
Compare
Choose a tag to compare

Revised release for submission for Udacity testing. Includes the following improvements:

  • Convert traffic light classifier model to frozen TensorFlow graph for faster initialization time to first classification.
  • Increased manual braking torque to 700 Nm to fully stop Carla according to Udacity engineer's feedback.
  • Add quick brake release filter reset for transition from 'full stop' to 'speed up' condition and optimize waypoint follower tuning to be able to quickly get through short green light cycles.
  • Add overshoot buffer for stop target waypoint to prevent Carla from rolling through a red light.

Revised release for submission FX-v2.1

06 May 03:33
6f35721
Compare
Choose a tag to compare

Revised release for submission for Udacity testing. Includes the following improvements:

  • Increased manual braking torque for fully stopping at very low speed.
  • Initialize TL state count to send actual light waypoint index as first message to Waypoint Updater so it can start with correct driving state considering the nearest light.

Second release for submission FX-v2.0

19 Apr 02:27
4732fb8
Compare
Choose a tag to compare

Second release for submission for Udacity testing. Includes the following improvements:

  • Prevent driving before traffic light classifier has finished initializing
  • Reduce min stopping distance to be able to stop at late-detected lights and improve calculation algorithm
  • Retune to stop closer to target stop line waypoint and release brake faster to get through short green light cycles
  • Improve traffic light classifier model with additional retraining
  • Improve traffic light state validation for yellow-red transitions
  • Slight tweaks to DBW parameter tuning for smoother control

First release for submission FX-v1.0

11 Apr 20:03
3a948a7
Compare
Choose a tag to compare

First release for submission for Udacity testing.

Second release candidate with issue fixes

11 Apr 03:19
Compare
Choose a tag to compare

Includes fixes for issues:

  • Warnings from get_min_stopping_distance() with max velocity increased to 60kph
  • Unknown->Yellow light state validation