Learning About Learning From A New Breed Of Cognitive Robot Print E-mail
Science - Science Enterprise
Written by TS-Si News Service   
Sunday, 06 April 2008 17:00
Learning About Learning From A New Breed Of Cognitive Robot.
Rockit by Herbie Hancock. Rockit is a hit single by Herbie Hancock from his 1983 album, Future Shock. It was written by Hancock, bass guitarist Bill Laswell and synthesizer/drum machine programmer Michael Beinhorn. The video, directed by Godley & Creme, was among the earliest to feature African Americans on MTV and includes Hancock as a keyboard player on television (closing shot).
Linköping, Sweden. People who become aware they might be born with Harry Benjamin Syndrome (HBS) depend on the findings of science and medicine, as well as the resolution of underlying issues that might impair their decision-making and potential treatment. Self-knowledge is a key consideration. The field of robotics now offers new insights into the nature and limits of human knowledge, an investigation that began with the systematizations of Plato and Aristotle in Ancient Greece.
 
Early Artificial Intelligence (AI) research gained note as an inquiry into the possibility of synthesizing what was known of intelligence at that time and positing machine equivalents. To accomplish such a task, it was necessary to found a new scientific paradigm. What are the rules for reasoning? Must intelligence always be based on a biological model? Is there an exclusive way to view the subject will it be an assemblage of approaches?
 
In 1973, Christopher Longuet-Higgins coined the term cognitive science in his commentary on the Lighthill report [1], which concerned - and often criticised — the then — current state of Artificial Intelligence (AI) research. Currently, cognitive scientists study the mind and intelligence, drawing their procedures and evidence from anthropology, biology, computer science, linguistics, neuroscience, philosophy, and psychology.
 
Or, perhaps something entirely new would be needed, an approach that could, in turn, inform studies of human and animal intelligence.
 
Metacognition.
Metacognition refers to higher order thinking that involves active control over the thinking processes involved in learning.
Activities such as planning howto approach a given learning task, monitoring comprehension, and evaluating progress toward task completion are metacognitive in nature.Because metacognition plays a critical role in successful learning, it is important for both students and teachers.
Both knowledge and strategy components are important.
Knowledge is considered metacognitive if it is actively used in a strategic manner to ensure that a goal is met. Metacognition is often referred to as "thinking about thinking" and can be used to help students “learn how to learn.”
Metacognitive knowledge involves executive monitoring processes directed at the acquisition of information about thinking processes. They involve decisions that help:
• identify the current task,
• check on current progress of that work,
• evaluate that progress, and
• predict the outcome of that progress.
Strategies are goal oriented. Cognitive strategies help to achieve a particular goal, while the metacognitive type are used to ensure that the goal has been reached.Both types involve executive regulation processes directed at regulating the course of thinking. They involve decisions that help:
• allocate resources to the current task,
• determine the order of steps to be taken to complete the task, and
• set the intensity or the speed at which one should work the task.
Cross-species occurrence. The ability to consciously think about thinking appears to be a unique characteristicof sapient species.There is some evidence that monkeys and apes can make accurate judgments about the strengths of their memories of fact.
However, attempts to demonstrate metacognition in birds have been inconclusive.A 2007 study provided some evidence for metacognition in rats.
Designers of artificial cognitive systems have tended to adopt one of two approaches to building robots that can think for themselves: classical rule-based artificial intelligence or artificial neural networks. Both have advantages and disadvantages, and combining the two offers the best of both worlds, say a team of European researchers who have developed a new breed of cognitive, learning robot that goes beyond the state of the art.
 
The researchers’ work brings together the two distinct but mutually supportive technologies that have been used to develop artificial cognitive systems (ACS) for different purposes. The classical AI approach relies on a rule-based system in which the designer largely supplies the knowledge and scene representations. The robot follows a decision-making process — much like climbing through the branches of a tree — toward a predefined response.
 
Biologically inspired artificial neural networks (ANNs), on the other hand, rely on processing continuous signals and a non-linear optimisation process to reach a response which, due to the lack of preset rules, requires developers to carefully balance the system constraints and its freedom to act autonomously.
 
Michael Felsberg, a researcher at the Computer Vision Laboratory of Linköping University in Sweden.“Developing systems in classical AI is essentially a top-down approach, whereas in ANN it is a bottom-up approach,” explains Michael Felsberg, a researcher at the Computer Vision Laboratory of Linköping University in Sweden [2]. “The problem is that, used individually, these systems have major shortcomings when it comes to developing advanced ACS architectures. ANN is too trivial to solve complex tasks, while classical AI cannot solve them if it has not been pre-programmed to do so.” Felsberg's project, Cognitive Systems Perception, Action, Learning (COSPAL) [3], is funded by the European Union (EU). 
 
Beyond the state of the art
 
Felsberg’s team found that using the two technologies together solves many of those issues. In what is possibly the most advanced example of such a system developed anywhere in the world, the researchers used ANN to handle the low-level functions based on the visual input their robots received and then employed classical AI on top of that in a supervisory function.
 
“In this way, we found it was possible for the robots to explore the world around them through direct interaction, create ways to act in it and then control their actions in accordance. This combines the advantages of classical AI, which is superior when it comes to functions akin to human rationality, and the advantages of ANN, which is superior at performing tasks for which humans would use their subconscious, things like basic motor skills and low-level cognitive tasks,” notes Felsberg.
 
The most important difference between the COSPAL approach and what had been the state of the art is that the researchers’ ACS is scalable. It is able to learn by itself and can solve increasingly complex tasks with no additional programming.
 
“There is a direct mapping from the visual precepts to performing the action,” Felsberg confirms. “With previous systems, if something in the environment changed that the low-level system was not programmed to recognise, it would give random responses but the supervising AI process would not realise anything was wrong. With our approach, the system realises something is different and if its actions do not result in success it tries something else,” the project coordinator explains.
 
“Like training a child or a puppy”
 
This trial-and-error learning approach was tested by making the COSPAL robot complete a shape-sorting puzzle, but without telling it what it had to do.This trial-and-error learning approach was tested by making the COSPAL robot complete a shape-sorting puzzle, but without telling it what it had to do. As it tried to fit pegs into holes it gradually learnt what would fit where, allowing it to complete the puzzle more quickly and accurately each time.
 
“After visual bootstrapping, the only human input was from an operator who had two buttons, one to tell the robot it was successful and another to tell it that it had made a mistake. It is much like training a child or a puppy,” Felsberg says.
 
Though a learning, cognitive robot of the kind developed in COSPAL constitutes an important leap forward toward the development of more autonomous robots, Felsberg says it will be some time before robots gain anything close to human cognition and intelligence, if they ever do.
 

In human terms, our robot is probably like a two or three year old child, and it will take a long time for the technology to progress into the equivalent of adulthood. I don’t think we will see it in our lifetimes.

Nonetheless, robots like those developed in COSPAL will undoubtedly start to play a greater role in our lives. The project partners are in the process of launching a follow-up project called Dynamic Interactive Perception-action Learning in Cognitive Systems (DIPLECS) [4]. The follow-on will test their ACS architecture in a car. It will be used to make the vehicle cognitive and aware of its surroundings, creating an artificial co-pilot to increase safety no matter the weather, road or traffic conditions.
 
“In the real world you need a system that is capable of adapting to unforeseen circumstances, and that is the greatest accomplishment of our ACS,” Felsberg notes.
 
 
 

[1] Artificial Intelligence: A General Survey. In: Artificial Intelligence: a paper symposium. Science Research Council. 1973.

 

[2] Michael Felsberg, is an assistant professor at Linköping University and a researcher in the Computer Vision Laboratory in Sweden.

[3] Cognitive systems using perception-action learning (COSPAL). Central topics of the COSPAL project are a new system architecture and new learning strategies for artificial cognitive systems (ACSs). The novelty of the approach lies in the interaction of continuous and symbolic perception and action, which results in robust and stable motor and sensorial capabilities of the system and allows a purposive behaviour. Moreover, the system is able to extend itself through exploration of the environment and has therefore the potential to solve problems of a large range of complexity levels. The new learning strategy is based on the assumption that perception is shaped by incremental (online) learning of percept-action mappings and by introducing rules at an appropriate level of tasks.

The necessary system structure to achieve this goal is a multi-level network consisting of different layers, using new kinds of networks and a new type of information representation, the channel representation, whose locality property allow a fast convergence in learning. In the demonstrator of the COSPAL project we will show that after an initial bootstrapping phase such a system can autonomously solve shape-sorter puzzles of various kinds.

[4] Dynamic Interactive Perception-action Learning in Cognitive Systems (DIPLECS). The DIPLECS project aims to design an Artificial Cognitive System capable of learning and adapting to respond in the everyday situations humans take for granted. The primary demonstration of its capability will be providing assistance and advice to the driver of a car. The system will learn by watching humans, how they act and react while driving, building models of their behaviour and predicting what a driver would do when presented with a specific driving scenario. The end goal of which is to provide a flexible cognitive system architecture demonstrated within the domain of a driver assistance system, thus potentially increasing future road safety.

In order to achieve these goals, the DIPLECS architecture must allow for learning and adaptation in dynamic, real-time and real-world scenarios. Starting from a basic, rudimentary capability, it must constantly refine and improve its capability by observing a human driver, the car data and the surrounding environment. The architecture applies a hierarchical design principle, where adjacent levels are connected by feedback-loops. Learning occurs in two ways, either by analysing human-car-environment interaction or by (cognitive) bootstrapping of its own capabilities.

 

This article is adapted and extended from materials provided by ICT Results in Europe, a service of Information and Communication Technologies (ICT).

 
TS-Si News ServiceThe TS-Si News Service is a collaborative effort by TS-Si.org editors, contributors, and corresponding institutions. The sources can include the cited individuals and organizations, as well as TS-Si.org staff contributions. Articles and news reports do not necessarily convey official positions of TS-Si, its partners, or affiliates. We welcome your comments. Use the form below to leave a public comment or send private correspondence via the TS-Si Contact Page. We will not divulge any personal details or place you on a mailing list without your permission.
 
Comments (0)Add Comment
feedSubscribe to this comment's feed
min/maxShow/Hide comments

Write comment
feedShow/Hide comment form

security code
Write the displayed characters


busy
< Prev   Next >
Last Updated on Monday, 07 April 2008 10:23