Skip to content

Weekly Reports

Jump to bottom
mbechtel2 edited this page Aug 3, 2017 · 23 revisions

6/15/2017

Last week

  • [James]
    • Task 1: Complete a URDF model of the car for simulating in RViz and Gazebo
    • Task 2: Set up a joint_state_publisher that uses Twist messages to move parts of the car
  • [Michael]
    • Task 1: Go through the video recorded from the car and save all frames as separate png images.
    • Task 2: Get an estimate of the steering angle and write it to a row in a csv file along with the path to the frame png.
  • [Nick]
    • Task 1: Set up development environment with ROS and tensorflow.
    • Task 2: Started build process for second car.

Plan for the next week

  • [James]
    • Task 1: Complete the joint_state_publisher so that the entire car can move within the RViz simulator
    • Task 2: Make sure that the RViz functionality is also present when running Gazebo
  • [Michael]
    • Task 1: Develop a Keras model that can be trained using the data obtained.
    • Task 2: If possible, load the recorded video onto the DeepTesla website to see if the model can also be trained there.
  • [Nick]
    • Task 1: Start testing if the deepdrive-universe model can be trained using the data that Michael collected.
    • Task 2: Setup second Pi and finish second car.

6/21/2017

Last week

  • [Michael]

    • Task 1: Change data collection to save all videos/data in separate dataset folders.
    • Task 2: Create a Keras model that is trained with data from all datasets.
    • Task 3: Train model using deeptesla (currently doesn't work for videos where initial frames don't exist).

  • [Nick]

    • Task 1: No longer exploring/experimenting with deepdrive or universe. Currently working on getting deeptesla to work on the pi.
    • Task 2: Waiting on motor hat and battery pack to finish second picar.

  • [James]

    • Task 1: Working on tf Broadcaster for RViz
    • Task 2: Converting message publisher into C++

Plans for next week

  • [Michael]

    • Task 1: Get all datasets/videos to work with deeptesla.
    • Task 2: Get data of car driving around and train the models with that data.

  • [Nick]

    • Task 1: Finish second picar.
    • Task 2: Work with Michael in collecting trainable data sets from the hallways.
    • Task 3: Get deeptesla working and training with our datasets.

  • [James]

    • Task 1: Publish odometry messages to robot model
    • Task 2: Determine angle for applying the transform broadcaster
    • Task 3: Update the ROS Tutorial Wiki page

6/29/17

Last Week

  • [James]
    • Task 1: Worked on TF angles and Odometry messages for carModel simulation
    • Task 2: Updated repository wiki for running using carModel simulation
  • [Michael]
    • Task 1: Captured initial data from picar and trained a model using it.
    • Task 2: Used deeptesla to create a visualization of model performance including statistics (error, average error, etc.).

Plans for Next Week

  • [James]
    • Task 1: Continue working on odometry message publishing
    • Task 2: Begin work on Gazebo navigation
  • [Michael]
    • Task 1: Collect more extensive training data that can be used to train the model.
    • Task 2: Modify deeptesla model to optimize performance and, potentially, reduce training time.
    • Task 3: Get visualization to work on the raspberry pi (currently not able to due to an opencv error).

7/5/17

Last Week

  • [James]
    • Task 1: Completed Odometry Publishing in RViz
  • [Michael]
    • Task 1: Create publisher node that takes predicted angles from the trained model using a video and publishes them to the cmd_vel topic. This got the wheels to turn in the RViz simulator.

Plans for Next Week

  • [James]
    • Task 1: Add camera/sensor for Gazebo model
    • Task 2: Work on OpenCV image conversion in ROS
  • [Michael]
    • Task 1: Get predicted angles from the model using images from the mjpg stream and publish those to cmd_vel.
    • Task 2: Test publishing messages to the picar and getting it to drive autonomously.

7/17/17

Last Week

  • [James]
    • Task 1: Worked on Video Converter and Stream Converter
  • [Michael]
    • Task 1: Made data collected in Udacity compatible with deeptesla training.
    • Task 2: Create ROS publisher and non_ROS controller that gets the picar to self-drive.

Plans for Next Week:

  • [James]
    • Task 1: Alter Car Controller for RViz publishing
    • Task 2: Research RViz Map-making
  • [Michael]
    • Task 1: Alter path and/or controllers to improve the car's self-driving.
    • Task 2: Create controller for Udacity for potentially testing models in that environment.

7/27/17

Last Week:

  • [James]
    • Task 1: Working on Stream and Video Driver
    • Task 2: Rviz Training for Datasets #1-4
  • [Michael]
    • Task 1: Added controller for testing the model in Udacity.
    • Task 2: Create ROS and non-ROS controllers that implement the DAgger method.

Plans for Next Week:

  • [James]
    • Task 1: RViz Stream Driver Mimicing the Actual Car
    • Task 2: Path Planning
    • Task 3: ROS Report
  • [Michael]
    • Task 1: Work on DAgger controllers to improve the car's self-driving.
    • Task 2: Find other sources for data collection.

8/03/17

Last Week:

  • [James]
    • Task 1: Updated Gazebo Car Model
    • Task 2: Researched path-planning and map-making for final report
  • [Michael]
    • Task 1: Added frame number and state variable value to video.
    • Task 2: Got video of the car driving on the track and in the hall.

Clone this wiki locally