Skip to content

Latest commit

 

History

History
85 lines (74 loc) · 3.09 KB

README.md

File metadata and controls

85 lines (74 loc) · 3.09 KB
Raw

metis

This package is contains tools for evaluating models and inference algorithms for task and motion planning. The name comes from Metis, a Greek goddess of wisdom, mother of Athena.

Maintainer

Dependencies

  • numpy
  • scipy
  • shapely
  • pybox2d
  • triangle
  • (optional) pydot
  • (optional) nosetests

Everything can be installed through pip.

pip install numpy scipy shapely Box2D triangle

If you are not in a virtualenv, this will install to /usr/ by default, and will require sudo privileges. If you'd rather install somewhere else, run

pip install --upgrade --install-option="--prefix=$MY_PREFIX" numpy scipy shapely Box2D triangle

with $MY_PREFIX set to, for instance, $HOME/rrg. Note that you must also update your PYTHONPATH environment variable accordingly to use the packages.

NOTE: installing the pybox2d from pip might fail if swig is not installed. If that happens, apt-get install swig may help.

Installation

This will provide minimal output. To run the tests and see the normal nose output, from the package directory run nosetests without arguments.

Usage

A simple block-pushing example is included; more comprehensive and extensible examples are planned for short term development in the future. This example world has a robot pushing objects between three rooms. To run the example, you must create a yaml description of the problem to solve. Examples of the format can be found in the cfg directory. Briefly, a valid job file specifies the start configuration of the world, as well as algorithmic parameters. For example:

noupper_blocked_forgg_0200_00: # name for this job
  algorithm: forgg             # the algorithm to use ('forgg' or 'tamp')
  domain:                      # Parameters affecting the problem domain
    upper_door: True           # True if there should be two doors between the
                               # rooms; False if there should be only one.
  instance:                    # Parameters affecting the problem instance
    box1:                      # The initial location of the first object
    - 3                        #  x coordinate (in meters)
    - 5                        #  y coordinate (in meters)
    - -0.2                     #  orientation (in radians)
    box2:                      # The initial location of the second object
    - 5
    - 2.5
    - 0.1
    robot:                     # The initial location of the robot
    - 2
    - 2
    - 0
  solver:                      # Solver parameters
    count: 200                 # Number of samples to generate in each
                               # mode factor
    seed: 0                    # PRNG seed value

To run this example, call

./run.py cfg/forgg.yaml

Command line output is sent to a log file by default, which you can view in real time with

tail -f noupper_blocked_forgg_0200_00.out

To solve the example problem with count=200, the planner must explre around 240000 vertices. On a typical desktop, the planner can explore around 200-400 vertices per second, for a total planning time of 10-20 minutes.