lms-org / config_auto_drive Goto Github PK

Evaluate it

Kamerakalibrierung mit opencv: http://docs.opencv.org/2.4/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html

Obstacle:
trust threshold dynamisch setzen
-Wenn man ein hinderniss erkennt wo keines ist erhöhen

• wenn man in eins fährt, erniedrigen

Berücksichtigen von Hindernissen auf der linken Spur vor oder nach Ausweichen

Fix auto-reconnect for socket_data

Crossings:

• detect crossings (may already work)
  • set position to the middle-lane
• search for obstacle crossing the crossing
• generate virtual points for the environment_predictor

Camera crashes, only reboot fixes it

nice approach: https://www.youtube.com/watch?v=opsmd5yuBF0

I added some #IMPORTANT, search the code to find and fix them

update camera settings automatically (via configsChanged() method)

Globale Maximumssuche nutzen und dann punkt mit linienbreite validieren

Implement ego-motion estimation (Kalman)

Send ip to website
http://superuser.com/questions/211411/executing-a-script-when-dhcpd-give-an-ip

Select driving mode

TODO:

• fix distance to obstacles
• fix distance to crossing

TODO:

• bounding-box for obstacles
• lines for crossings

use start line to measure the lap time

Start lms automatically on boot,
use symlink

Reset data on failure (driving with rc)

Straßenmodell:

• Umfeldmodell verschieben führt zu komischen ergebnissen
• Sensorrechtecke anpassen (hinderniss blieb außerhalb stehen)
• ego-estimater, error erkennen und dann die rate-methode benutzen

Automatically download/compile/install config_repos,
Support more then one car in: http://car.tum-phoenix.de/

TODO

• Move obstacles with ego motion
• (Detect if obstacle is moving)

Threshold für die Bilderkennung:

• sobelThresholds ermitteln
  Helligkeits-Histogramm für kleinen Bereich vor dem Auto erstellen

Doesn't trust small obstacles on the side of the road

Set breaking light

Some of their approaches might be useful for us: https://sites.google.com/site/orcaracer/previous-projects/completed-projects

The problem is, that I don't think that we have enough computational power.

Those heavy-lifting operations could run parallel and be used to improve our results. (Not sure if it's worth it)

Doesn't work properly

only dirs are created, no files are written

add more rgb light-effects

Test current state, improve it

calculate localDx and localDy (TODO)

Make sure that ueye is running, block until a valid image was received:

• wait till device is available
• restart ueye driver
• count failed images

Add:

• motor-sound (depending on speed)
• radio (playlists)

https://github.com/tcbrindle/sdl2-cmake-scripts

Count infringements: https://github.com/tum-phoenix/config_auto_drive/tree/master/external/modules/infringement_counter

Trust:

• move trust to StreetObject
• trust should become a float from 0 to 1
• handle trust merging (occurs if two objects are merged)

Set indicator (turn signal)

Add ipv6 support to socket-connection

Add some basic parking algorithm

Some problems are solved using dirty hacks

Handle reset-button

Send Obstacles via socketData using cereal. The last time we tried multiple inheritance wasn't working

Add some working speed-controller

add race-mode:

• calculate optimal trajectory
  -ignore middle-line

Improve obstacle handling/detection

Improve crossing viewDir

Damit könnte man parallel Ausweichen, in der Kurve nach innen schauen etc.

I would like to record the CONTROL_COMMANDs and replay them.