MRG icon
Chris Adams

MOBILE ROBOTICS RESEARCH GROUP

research:

learning by imitation

 

biological models

social robots

learning by imitation

artificial life

 

Contact: Yuval Marom
Last updated: Wed Apr 10 10:01:27 2002

Current researchers

Previous researchers


Learning by Imitation

We do not exist alone. Humans and most other animal species live in societies where the behaviour of an individual influences and is influenced by other members of the society. Within societies, an individual learns not only through classical conditioning and reinforcement, but in large extent through observation and imitation. We are concerned with how to add such learning mechanisms to artificial agents, including simulated and mobile robots.

We are investigating the following:

The work that started this project off involved experiments on how to make a robot observe, imitate, and learn from another robot or human. This work was inspired by several fields including developmental psychology, biology, and neurophysiology.

For more detail on imitative learning mechanisms, visit John Demiris's and George Maistros' homepages.

From this early work emerged the question of when should an agent decide to imitate? under what conditions is the presented stimulus worth further processing, eventually ending up in learning? This could be treated as an issue of attention. At least two factors are involved here:

  • internal motivations and needs
  • influence from the social environment
For more detail see below or visit Yuval Marom's homepage.
 

Attentional Mechanisms for Imitative Learning

In social learning, an individual benefits from interacting with its social environment, to acquire new competencies and skills. In other words, the existence of one or more other individuals in its perceived environment, aids an individual to learn as it negotiates unknown environments. Psychologists use the term "social enhancement" to refer to all social influences on an individual's performance. One form of such influence is "local enhancement" or "stimulus enhancement", where one (or more) individual - the teacher - actively manipulates the perceived environment of another - the learner, either by moving objects in the environment, or changing its own location, configuration, orientation etc'.

The purpose of these manipulations is to direct the attention of the learner to the relevant stimuli of the task to be learned. In learning by imitation, a student agent learns by imitating a teacher agent, and then associating its perceived environment with the motor commands that it had to execute in order to imitate. The next time the student finds itself in a similar situation, it can execute the commands it has associated, and hence perform the correct actions. Hopefully, by having the student's attention focused on the relevant features of the environment associated with the task at hand, and at the right time, the amount of storage required for representation and memory is reduced, and therefore the speed of learning is increased.

We are interested in modelling perception of change that is necessary to determine when it is worthwhile (for the agent) to pass information from perception, on to further processing and storage in memory, ie learning. We do this by providing our agent with a concept of significance of stimuli, and designing a strategy whereby the agent can adapt its own criterion for the significance of stimuli that will be worth further processing, based on built-in attentional capacity. See the following papers for more detail:

To summarise, we are investigating how what we call "attention", together with social interactions, can help a mobile robot learn from other agents, robotic or human.

For more papers, visit Yuval Marom's homepage.


 

Learning to Communicate

In previous work, a set of experiments investigated learning to communicate through imitation. See