Adaptive Cognitive Systems

Cognitive Modeling Research and Development

Table of dTank Demos

The table below describes a series of demonstrations we've developed on this project that show both the good and the bad of the approach we used.  The behavior that is expected is in the description field, while extra information about the demo and how it was created is in the notes field. If the name of the demo is clickable, it will take you to the associated demo video!

dTank Demo Status
Demo NicknameDateStatusDescriptionNotes
Aim10/16/09GoodLocates and drives to an officer; stops when reaches the officerTo try different officer starting positions, you must manually edit the file
Chase-Short 110/16/09GoodFollows a bot or human tank around at a set distanceThis version uses opponent orientation information as an attribute, and so is able to stay at the distance easier when the opponent is coming at it.
Chase-Short 210/16/09GoodFollows a bot or human tank around at a set distanceThis version doesn't use opponent orientation, so is a bit jerkier.  Also does not follow a human around as well.
Chase-Long10/16/09FairFollows a bot or human tank around at a set distanceThe instances for Chase-long were 600 seconds of recording, so there is much more following, and not as much finding/looking for the tank. This means that the model can only chase tanks that come into view on their own.  if the starting point of the blue tank is moved on the chase-flat map, the model works, but not very well.  This is a case of bad instances, not a bad set of attributes.
Draw-Easy10/16/09GoodDraws an "N" and then circles the map to the left, finishes by going up and down the left side of the mapThis demo only uses the time stamp as an attribute. It is an example of how CIBRE can be used to instantly generate scripted behavior. There is no flexibility or generalization here, however.
Path10/16/09GoodNavigates the path through the woods (sometimes gets lost in the circle towards the end)This model preloads the map.  It also must be run in real-time, because of the overhead concerned with annotating and updating the entire map.  It may continue around the circle because it has no knowledge as to where the end of the path is (it is a steering model, not a path model).
Path-Untrained10/16/09GoodNavigates the path through the woodsThis is the exact same model as Path, right down to the instances. However, it is an awesome demo because our SME never drove this path! It shows how flexible an instance-based model can be.
Pong10/16/09FairPongs ("patrols") between wallsThis demo was created on-the-fly in front of a customer during the site visit at the end of Phase I. There are several flavors to it. In the pong.cfg file, you can choose between three maps: pong, pong-diag, and pong-vert. You can also choose to preload these maps (:map 'pong preloads the pong map). All of the maps work when pre-loaded, but the model makes some odd decisions when the map is not preloaded.
Target-Officer10/16/09FairLocates an officer, and then shoots itThis model, due to the driver's rigidity, may miss the officer if it is not located at an even angle (160, 150, etc).  This model occasionally works on tanks, but will often quit responding if the tank manages to move out of sight when the model is not yet loaded.