(Expert in Information Science) Lecture

  • Artificial intelligence: Deep Learning, Machine Learning, etc.
  • Traditional robots and traditional artificial intelligence
    • Robots are about the real world (physical models, linear approximations) (electricity and machinery)
    • Artificial intelligence is about the formal world (graphical models) (information and communication)
    • In the past, these were quite separate fields.
  • However, they are gradually merging.
  • Example: Application of deep learning (Computer Vision) to robots
    • There is a dataset of images (RGB-D) and their corresponding grip positions.
      • RGB-D images include depth information in addition to RGB.
    • By learning from this dataset, the robot can determine the grip position of what it sees.
    • (By the way) Experiment: Which information in RGB-D is important?
      • When each piece of information in RGB-D was removed, RGB was found to be more important than depth (removing RGB made the image black and white).
    • However, even when using this dataset, the robot was unable to actually grasp objects (!).
    • It is not enough to rely solely on visual perception and target positions; the robot’s Embodied Knowledge is important.
      • To be able to grasp objects, it is necessary to capture the aspect of the robot’s actual body interacting with the environment.