mpavezb/CarND-Capstone

The final project for Udacity Self-Driving Car Nanodegree.

This is the project repo for the final project of the Udacity Self-Driving Car Nanodegree: Programming a Real Self-Driving Car. For more information about the project, see the project introduction here.

Team Members

• Georgios Fagogenis, georgios.fagogenis@tttech-auto.com
• Matias Fernando Pavez Bahamondes, matias.pavez-bahamondes@tttech-auto.com (Team Lead)
• Natalia Estefanía Jurado Espín, natalia.jurado@tttech-auto.com

Resources

• Installation: Local/Docker setup, simulator, and dependencies.
• Testing: How to test on the simulator and the real car data.
• Implementation: Design and implementation details.
• Results.
• Acknowledgements.

Testing

Build

Compile the project locally or on the docker container:

cd ros
source /opt/ros/kinetic/setup.bash
catkin_make
source devel/setup.sh

Simulator

1. Run the simulator locally: The camera must be enabled and the manual mode disabled.
2. Launch styx.launch locally or on the docker container.

cd ros
roslaunch launch/styx.launch

Real World Data

1. Run the Rosbag file that was recorded on the Udacity self-driving car:
2. Launch the project in your project in site mode

rosbag play -l traffic_light_bag_file/traffic_light_training.bag

cd ros
roslaunch launch/site.launch

Implementation

Waypoints

• TODO: About decceleration profile for the trajectory.

DBW Controller

• TODO: How the PID for throttle was tuned.
• TODO: Insights on the yaw controller.
• TODO: How to deal with stop and go scenarios.

Trafic Light Detector

Labeled data, and training scripts for the traffic light detector were provided by user vatsl on the repository TrafficLight_Detection-TensorFlowAPI.

A different container matching the exact dependencies as in the real car was used for training the model. Instructions can be found in the respective training section.

• TODO: About the dataset.
• TODO: About the model.
• TODO: About the results for simulation.
• TODO: About the results for real car data.

Results

• TODO: Add images/video.
• TODO: (Improvement - Corner case): About how dealing with yellow lights would improve the stop.

Acknowledgements

• Most of the implementation for ROS nodes and controllers was obtained from Udacity classroom lessons.
• Labeled data, and training scripts for the traffic light detector were provided by user vatsl on the repository TrafficLight_Detection-TensorFlowAPI.