Robotic Tasks in Theo

• Internal generation of goals

  In Theo-Agent, goals are created within the agent, based on pre-specified patterns of behavior called drives and the current world state. The drives simply specify the conditions under which a particular goal should be activated and the conditions for which it may be considered achieved. Theo-Agent also contains rules for prioritizing goals should several be active at once, thus allowing it a rudimentary capability for considering multiple, simultaneous goals.

• Caching policy.

  All inferred beliefs are based directly or indirectly on sensor input. When sensor inputs are lost, the cached inferences about the inputs are deleted by the architecture's simple truth maintenance system and are only re-derived if needed. Thus, for example, Theo-Agent may only be aware of objects in the world if they are currently being directly sensed or if information about the objects are part of the formation of some plan. In other words, Theo-Agent has little ability to rationalize object permanence.

• Control Policy

  Theo-Agent executes an infinite loop consisting of the following steps:
  1. Sense
  2. Decide
  3. Execute
  The decide portion of the loop is responsible for retrieving current beliefs about actions, inferring a chosen action from the available S-R rules, or invoking some applicable plan. There is also a default response to WAIT but it is unclear when this response is invoked and when the impasses due to a lack of knowledge drive further deliberation (e.g., plan generation).

• Sensing Policy.

  Theo-Agent uses a sensing strategy to avoid the necessity of sensing everything in knowledge-rich environments.

Go to a discussion of this capability for multiple architectures.

Current Location: Theo-Capabilities-Planning

Go to NEXT page.