Knowledge Representation and Ontologies for Autonomous Systems Symposium

Break-Out Group Challenge Problem

March 22-24, 2004

Scenario:

A team of five autonomous robots is tasked to perform trash removal for an airport. This involves identifying and removing trash from the floor and seating area. The airport corridors contain lanes (marked by yellow tape) that the robots must follow when looking for trash. Each lane is bidirectional, though the lane is only able to fit one robot at any given location. The robots can only leave the lanes when observing or picking up a piece of trash that is not in the lane. People in the airport are not restricted from entering or leaving the lanes.

The robot must be able to distinguish between people, trash, and other robots. For each piece of trash that is identified, the robot must keep a record of all pertinent information about that object and why it classified it as trash.

The robots are also expected to recycle, when possible. There are specially marked trash receptacles around the airport for paper, glass, and plastic recycling. There are also numerous trash receptacles for all other forms of trash. The robot must identify the type of trash it finds and place it is the proper receptacle.

All of the robots must cover the entire airport multiple times every day. The robots should know where other robots are at all times, and coordinate their activities to ensure that they remove trash from the airport in the most efficient manner possible.

The robots must have full awareness of their own health status and know what action they are performing at all times. They must be able to communicate this information with the other robots and proceed to a repair facility when something goes wrong.

The robots need to be cognizant of people who are walking through the hallways. The robots should be no closer than a predetermined number of feet from any person at any time, for safety considerations. Hence, the robot must make near-term predictions as to where any person will be at points in the future to ensure that it will never be closer than allowed.

While doing trash clean-up, the robots are also tasked with identifying any suspicious package that they encounter, based upon qualities such as its location, its color, its odor, its shape, and its size. When a robot encounters a suspicious package, it has a series of steps it must follow, including not getting too close to it, assessing its criticality, and providing an emergency alert to a central base. It must also immediately inform the other robots as to the location of the package.

Each robot is equipped with:

  • A location system which informs it of its approximate location at all times
  • A sensory processing system that allows it to:
  • Segment objects in the environment
  • Identify the color, shape, size, and odor of objects in the environment
  • Know the exact location of objects in the environment
  • Identify lane markings in the corridors
  • A communication system that allows it to:
  • Transmit information to other robots
  • Receive information from other robots
  • Separate finite-capacity holding containers for glass, plastic, paper, and all other forms of trash
  • A robot arm that can:
  • Grasp any type of object and place it in the proper holding container
  • Remove a bag from the appropriate holding container and place it in the appropriate trash receptacle
  • Re-bag each holding container

A mobility system that can:

  • Move the robot in any direction

Your task:

Define a knowledge architecture that provides a mechanism to capture all pertinent information for the robot to perform its duties. Within the architecture, you should include the different types of knowledge that will be included, how that knowledge will be represented, and what type of interfaces will be needed between the knowledge sources. In addition, you should be able to address the questions below in your architecture. Note that you are not designing the robot itself, you are only designing the “world model” of the robot, and how that world model information will be used by the robot to perform its duties.

  • What types of knowledge are needed for the autonomous system to behave properly?
  • Spatial knowledge?
  • Rule-based knowledge?
  • Symbolic knowledge?
  • Parametric knowledge?
  • A priori knowledge?
  • In situ knowledge?
  • Self-awareness knowledge?
  • History knowledge?
  • Intent knowledge?
  • Others?
  • What formalisms/approaches should be used to capture that information?
  • How will various formalisms be integrated into a unifying structure?
  • What role do ontologies play in this scenario?
  • What are the real-time considerations, and will the knowledge representation approaches address those requirements?
  • What are the inference requirements? Will various forms of inference be required? If so, how will they all be integrated?
  • How do we ensure the quality of the knowledge in the knowledge base?
  • What are the components the autonomous robot that need to use the knowledge, what pieces of knowledge do each need to use, and what the interfaces that each need to access the information?
  • What role, if any, will learning play in this system?
  • How should uncertainty be captured in this scenario?