ANIMATION FROM NATURAL LANGUAGE (ANIMNL)

AnimNL (http://www.cis.upenn/edu/ mmoore/animnl.htm) is being developed by two research areas at the university of Pennsylvania; the Centre for Human Modelling and Simulation (HMS) and the Language, Information and Computation (LINC) Laboratory.

It's goals are to provide realistic animation of a human figure carrying out tasks as a demonstration of the understanding of the natural language instructions given it.

The focus of the research has been on using instruction manuals for the natural language tasks. These are good examples of instructions because although they assume the user has little knowledge in the specific application or task they set out to describe, they do assume world knowledge, with which the user should be able to interpret the commands and carry them out.

The project aims to expand the current goals in the area of NLU research by not only taking chunks of more than one sentence at a time to parse and understand, but also by looking at allowing the agent to evolve it's understanding of a task by carrying that task out. The following two situations are given as examples:

• When an agent is told to ``Go into the kitchen to get the coffee urn'', he does not need to ground the definite expression the coffee urn before he begins to act. All that is required is that he be able to establish a referent once he gets to the kitchen. The understanding process must be able to allow for this delay.
• When an agent is told ``Vacuum against the direction of the pile to leave it raised'', the agent can find out through vacuuming what direction of sweep leaves the carpet pile raised. Again, he does not need to know the referent before starting to act, but he must be able to use the instruction to guide what it is he needs to know.

AnimNL uses model-based simulation provided by a system called Jack (TM) developed by the University of Pennsylvania's Computer Graphics Research Laboratory. Jack provides realistic modelling of human models, with reasonable mechanics, and many behaviours such as walking, stepping, looking, reaching, turning, grasping, strength-based lifting, and collision-avoidance posture planning.

The system is designed so that varied environments can easily be created and modified enabling designers to vary the situations in which the agents are placed.