locked
Autonomous driving using 3D LRF RRS feed

All replies

  • Great job! Congrats on building this simulation.

    Trevor

     

    Wednesday, December 7, 2011 9:40 PM
  • Did you use GPS ? How are you identifying the road and how are you smoothly taking the curves with out overshooting the bends.
    George Peter
    Saturday, December 10, 2011 12:45 PM
  • I didn't use explicitly the GPS sensor that comes with MRDS. However, I'm polling for the coordinates and orientation of the vehicle frequently.

    The user can specify a list of waypoints on the map before the simulation starts, which should help guide the vehicle until the destination point. In the video that I uploaded each two consecutive waypoints are chosen to be in a range between 50 and 100 meters.

    Every time a new point cloud is acquired through the 3D LRF sensor mounted on top of the car, I perform terrain classification to determine which points are obstacles and which points are safe to drive. The algorithm for now is straightforward and uses slope and distance data in each vertical plane scanned to determine if two consecutive points are within a predefined angle and distance bound considered acceptable.

    Safe points appear in green and obstacles appear in red. There are about 35000 impact points found on each scan, and the scan itself lasts around 110ms. That's an image from the simulation in which I tried to overlay the point cloud data.

    Regarding path finding, I’m using a similar idea to harmonic potential fields to which I apply A*. Then I make sure the vehicle follows the path as rigorously as possible. The path appears as a blue line on the navigation panel.

    There are many more improvements I’d have to make in the near future. You can see for instance in 02:45 how the vehicle gets off the road because it considers another path.

    Saturday, December 10, 2011 3:31 PM
  • That's good to know and very very interesting . Will you be sharing the code in codeplex  projects. I would love to use your simulation environment . I was trying to do this for a long time , i was not very successful .
    George Peter
    Saturday, December 10, 2011 5:25 PM
  • I would really like to share the entire project. However, parts of it are going to be contained in an Imagine Cup project and it's a matter of discussion with my team if they can be disclosed.

    I will gladly upload the simulated 3D Laser Range Finder service in a week or two in codeplex. I would like to know if the Robotics team is planning to have such a sensor incorporated in MRDS soon.

    Sunday, December 11, 2011 12:34 AM
  • There are no plans for a 3D Laser Range Finder. We are concentrating on Kinect. However, Kinect could not be used in this application because it does not work outdoors and does not have the Field of View or range that you require. So we are looking at indoor environments only, hence the Robotics@Home competition.

    Trevor

     

    Sunday, December 11, 2011 10:08 PM