Next steps

Collect data and test training

Data

• Mount sensors to robot
• Record rosbags to train model
• Figure out how much data this model trained on (on spreadsheet)

Train

• Train terrain representation model
• Visualize
• Train terrain preference cost model
• Visualize

Generate costmap

ROS node that handles the image processing, model inference, and costmap publishing.

• Imports the PyTorch model, takes in a patch and returns a scalar cost image