SocioCognitiveRobotics/CognitiveRobotics

From SocioCognitiveRobotics
Jump to: navigation, search

Cognitive Robotics

Cognitive Robot Architectures

A Cognitive Robot Architecture (CRA) is a robot control architecture that supports and implements various capabilities or functions such as planning, reasoning, and maintaining a mental state (consisting of e.g. beliefs, knowledge, goals). In cognitive science, such capabilities are called cognitive functions.

Note that an CRA does not need to be based on a cognitive architecture. Other approaches may be used and may be more successful for engineering cognitive capabilities for robots, i.e. to develop robot cognition. This means that the cognitive functions supported by an CRA do not have to faithfully model the same human cognitive functions. In other words, it is not obvious that a model of human cognition is required for developing robot cognition. See below for approaches that do use cognitive architectures for developing robot cognition.

Cognitive Robot Architectures based on Cognitive Architectures

Cognitive architectures aim to provide a computational model of human cognition. See for a partial overview and analysis of various cognitive architecture projects [1]; Soar, ACT-R and ADAPT have been applied to robotics. The main motivation to adapt and apply existing cognitive architectures to control robots is to create a robot that has similar cognitive functions and is able to apply these to solve complex tasks. Cognitive architecture seem to provide a logical starting point to this end. The idea is to transfer the cognitive functions present in such architectures to a cognitive robot control architecture.

Soar

Soar has been applied to robotics [REF].

Motivation:

Description:

Evaluation:

ACT-R

ACT-R has been applied to robotics [REF]

Motivation:

Description:

Evaluation:


ADAPT

The ADAPT architecture is based on the Soar cognitive architecture [2]. ADAPT is an acronym for Adaptive Dynamics and Active Perception for Thought. It is said to resemble in many respects the Soar and ACT-R cognitive architectures but is claimed to generalize the search control mechanisms of Soar and ACT-R (Benjamin, Lyons, Lonsdale 2004). The development and adaptation of the ADAPT architecture for robot control is a joint effort of Pace University and Fordham University which started in 2004. Work on the ADAPT architecture seems to have been started already in 1993 (Lyons et al 1993).

Motivation: The main motivation is to provide robots with a sophisticated range of behaviors [sic] including the use of natural language, speech recognition, visual understanding, problem solving and learning that can be used to handle complex analog environments in real time. ADAPT is also intended as a tool to explore the integration of perception, problem solving and natural language at a deeper structural level (Benjamin, Lyons, Lonsdale, 2004).

The motivation to create a new cognitive architecture, i.e. the ADAPT architecture, has been that cognitive architectures "do not easily support certain paradigms of perception and control that are mainstream in robotics."

Description:

Evaluation:

References

  • Benjamin, D.P., Lyons, D.,Lonsdale, D. (2004) ADAPT: A Cognitive Architecture for Robotics. 2004 International Conference on Cognitive Modeling.
  • Lyons, D.M., Hendriks, A.J., Shrivastava, S., Kallis, A.D. (1993) The ADAPT User Manual, Philips Research Tech. Report TR-93-013.