This is the spatial language understanding package written by Stefanie
Tellex, Thomas Kollar, with many other contributions.

* You must install java and set the JAVA_HOME environment variable.
* Run "make" to compile 3rd party dependencies. 


**************************
Forklift Demonstration

* To test the forklift mobile-manipulation code, run "rake
  train_forklift_aaai11" in crf_training to build the model.  Then run
  "rake tests" in forklift.  If all these test cases pass, it is
  successfully following commands such as "put the tire pallet on the
  truck."

* Then run "rake cost_function_browser" in forklift.  This command
  creates the the debugging interface for the inference.  You can type
  commands into the "Input Command" box.  Click "submit" to run the
  inference.  The parsed command appears in the parsed command box,
  and scored, inferred plans appear in the "Inferred Plans"box.
  Finally a graphical representation of the scene and inferred plans
  appears in the bottom pane. If you click "Windows/Context 3d" you
  can see a 3d representation of the scene and play back the inferred
  plan using the slider.  If you select a consituent in the "Parsed
  Command" window, it will be highlighted in the 2d and 3d views of
  the inferred action sequence.

  The system does not just infer a single plan in response to a
  command. Instead it tries many different groundings and returns a
  list of candidate assignments, with a score.  Each candidate
  assignment of grounding to linguistic constituents has a row in
  "Inferred Plans."  If you select a different row, you can see
  groundings associated with each inferred plan in the list.  Since it
  does approximate beam search, you won't see all possible
  combinations (depending on the beam width and the size of the
  environment.)