FAQ

What happens if an object falls to the ground?

Points will be deducted if a can or bottle falls onto the floor, or into the wrong bin. Non scoring items such as cardboard boxes will not give or lose you any points.

How will submissions be scored?

Each submission will be graded on a gazebo world that is not available to participants. A rubric will be used to score the model in the submission on this world. You can find the rubric and more information on scoring in the challenge guide located on the challenge page.

How can I pace a Simulink model to run in real time?

To set the pace of a Simulink model (relative to a real clock) follow these steps:

  1. Open your Simulink model

  2. In the Simulation tab, select Run > Simulation Pacing

  3. Select ‘Enable pacing to slow down simulation’

  4. Select the desired pace at which the model should run

For more information on pacing a Simulink model, refer to this documentation page.

How do I sync Simulink execution with Gazebo?

To sync Gazebo and Simulink so they run at the same time step, use the Gazebo Pacer block found in the Simulink model of the template files provided. To use this block, you will need to manually add the gazebo plugin to the VM. For instructions on setting up Gazebo for co-simulation, refer to this documentation page.

What sensors are available in the Kinova manipulator?

You will have access to an RGB camera, and a depth sensor (point cloud). These are the only sensors that you should use for obtaining information on the environment, and building your models. Starter code for using these sensors can be found in the template code found on GitHub.

How do I process point clouds in MATLAB from the depth sensor?

The depth sensor provides both images and point clouds that you can work with. To access the point cloud data, subscribe to the topic: /camera/depth/points. This will provide the XYZ data of the point cloud data from the camera. To learn how to work with this point cloud data, check out this video.

How can I use deep learning for object classification?

There are 2 main steps for training a deep learning model for object detection. Preprocessing the data and training a network using that data. With MATLAB, you can use built in apps to avoid doing this work from scratch. Check out these videos showing how to use built in features to build an object detection model:

  1. Preprocessing data

  2. Using a pretrained network

  3. Design and train a Yolov2 network

For more information on object detection, check out the documentation page.