mxwg / carnd-capstone Goto Github PK

Apparently at one location a red light is detected where there is no light at all. See the attached screenshot for the location and the (apparently) wrongly classified image. Perhaps we can add the image to the negative training set.

The waypoint updater node should adapt the velocity when it is aware of an approaching traffic light that is red.

It seems to work on the highway, but we should make sure it works on the parking lot as well (esp. after we have the detector).

Get the car to move around, no sophisticated control yet.

This state can be used to generate classified images or subbed into your solution to help you work on another single component of the node. The state component of the topic won't be available when running your solution in real life so don't rely on it in the final submission.

The waypoint updater node should set some velocity based on how fast we want the car to go (e.g. as close to the speed limit as possible) and should slow down in front of red lights.

Use a PID controller for velocity control instead of sending a fixed throttle command.

I could not see that any images were classified when playing back a bagfile.
I tried with rosbag play -l traffic_light_training.bag and was looking on the /traffic_waypoint topic for detected lights (nothing was published because there was not nearest light).
Adapting the code slightly so that the classification was always called seemed to show that red lights were detected often (even though they were not in the image) and other colors were not recognized at all.

We need to have some instructions on how to tell if the classification on such a bagfile is working.

The traffic lights from topic /vehicle/traffic_lights should be visualized at their correct locations with their current status.

The waypoint visualization could show the velocity of the upcoming nodes by adapting the size and/or color of the arrows proportionately to the velocity.

Implement the function get_closest_waypoint and return the closest waypoint.

When running with styx_site.launch and the parking lot simulation, the first waypoint seems to be around index 26. When letting the car run from there, it runs until around index 3 and stops.

I think the expectation is that it goes on until shortly before the starting point, but this is up for discussion.

We should have a convenient lauch file for the second track as well (styx_site.launch instead of styx.launch).

This will help in tuning the control for this different environment.