Cognitive robotics

From Wikipedia, the free encyclopedia

Cognitive robotics (CR) is concerned with endowing robots with mammalian and human-like cognitive capabilities to enable the achievement of complex goals in complex environments. Cognitive robotics is focused on using animal cognition as a starting point for the development of robotic computational algorithms, as opposed to more traditional Artificial Intelligence techniques, which may or may not draw upon mammalian and human cognition as an inspiration for algorithm development. Robotic cognitive capabilities include perception processing, attention allocation, anticipation, planning, reasoning about other agents, and perhaps reasoning about their own mental states. Robotic cognition embodies the behaviour of intelligent agents in the physical world (or a virtual world, in the case of simulated CR). In the most ambitious version, this implies that the robot must also be able to act in this real world.

Hence, a cognitive robot should exhibit:

knowledge
beliefs
preferences
goals
informational attitudes
motivational attitudes (observing, communicating, revising beliefs, planning)
capabilities to move in the physical world, and to interact safely with objects in that world, including manipulation of these objects

Cognitive robotics involves the application and integration of various artificial intelligence disciplines but is primarily inspired by psychology and brain science research. On the other hand, the robot capabilities will be limited by the current state of the art in robotics: the mechanics and electronics of robots are still very inferior to what humans have available, especially in the areas of tactile and visual sensing, the smoothness and energy efficiency of motion (including walking and object manipulation with fingers), and the task-directed planning of actions.

Core topics include knowledge representation, motivation, automated reasoning, planning and learning. A number of different methodologies can be adopted within cognitive robotics. These methodologies include not only the approach of classical symbolic AI —emphasizing symbolic reasoning and representation— but also more biologically-inspired approaches that use noisy and distributed representations of knowledge. One approach that attempts to merge a symbolic approach with a connectionist approash is SS-RICS. More purely connectionist and dynamic systems approaches for instance include Continuous Time Recurrent Neural Networks (CTRNNs) as studied by Randall Beer and colleagues and Adaptive Resonance Theory (ART), developed by Stephen Grossberg and colleagues.

One of the learning techniques that are used for robots is learning by imitation: the robot, provided with all the sensors and physical hardware needed to perform a human task, is monitoring the human performing a task, and then the robot tries to imitate the same movements that the human performed in order to achieve the task. Using its sensors, the robot should be able to create a three-dimensional image of the environment, and to recognize the objects in that image. A major challenge is hence to interpret the scene, and to understand what objects are needed in the task and which are not.

A more complex learning approach is autonomous knowledge acquisition: the robot now uses its sensors and its knowledge about the physical properties of the world, and is then left to explore the environment on its own. One of the terminologies of this behavior is called motor babbling. Basically the whole idea of this approach is to let the robot discover its capabilities on its own.

Some researchers in cognitive robotics have begun using architectures such as (ACT-R and Soar (cognitive architecture)) as a basis of their cognitive robotics programs. These architectures have been successfully used to simulate operator performance and human performance when modelling laboratory data. The idea is to extend these architectures to handle real-world sensory input as that input continuously unfolds through time.

Some of the fundamental questions to still be answered in cognitive robotics are:

How much human programming should or can be involved to support the learning processes?
How can one quantify progress? Some of the adopted ways is the reward and punishment. But what kind of reward and what kind of punishment? In humans, when teaching a little infant for example, the reward would be a chocolate or some encouragement, and the punishment will have many ways. But what is the effective way with robots?