Skip to content

Latest commit

 

History

History
153 lines (110 loc) · 4.51 KB

tom.md

File metadata and controls

153 lines (110 loc) · 4.51 KB

ROS, Planning, and TOM

The TOM examples require the TOM Robot package to execute.

  • Installing Dependencies
  • Real Robot Setup
  • Task Parser
  • One-Arm Orange Tests

Installing Dependencies

When using the real robot, we assume that you have access to a working version of the full CoSTAR stack. This contains perception code, data collection tools, et cetera, for running a robot and using our custom UI and other tools. You will not need all of this for TOM, so certain features can be freely disabled.

There is a prototype installation script, install_tom.sh, which should install parts of the CoSTAR stack and also CoSTAR plan.

Disabled Packages

  • costar_gazebo_plugins does not support kinetic/Gazebo 7
  • roboticsgroup_gazebo_plugins does not support Gazebo 7
  • robotiq_s_model_articulated_gazebo_plugins does not support Gazebo 7
  • sp_segmenter requires opencv 2

Real Robot Setup

See the guide to getting started with the real TOM robot

There are a few different pieces that need to start up:

  • the planning context
  • the perception backend, including the Alvar AR tracker
  • the planner itself

You can start up the context and perception systems with the command:

roslaunch ctp_tom planner.launch

By default, the real:=true option is set, meaning it will start up a perception pipeline. We also plan on supporting a fake version which creates a scene based on fixed TF poses. This does not simulate object interactions or anything fancy like that.

Important other launch files (included by default):

  • scene.launch: loads object descriptions used to create MoveIt planning scene
  • perception.launch: used to start real robot or Gazebo perception

For more information, see the real TOM guide.

The Task Parser

The ctp_tom parser is a version of the task plan parsing tool that takes in messages and produces an executable task graph. You can feed the parser one or more rosbags:

# One bag only
rosrun ctp_tom parse.py --bagfile oranges_2017-12-13-19-01-15.bag
# Multiple bags
rosrun ctp_tom parse.py --bagfile oranges_2017-12-13-19-01-15.bag,blocks_2017-12-13-20-07-27.bag

Adding the --fake flag will add some fake objects and compile the model:

rosrun ctp_tom parse.py oranges_2017-12-13-19-01-15.bag,blocks_2017-12-13-20-07-27.bag --fake

You can specify the project directory to which the resulting models should be saved with the --project $NAME flags:

rosrun ctp_tom parse.py --bagfile oranges_2017-12-13-19-01-15.bag --fake --project oranges

Running With Fake Objects

The fake_objects.py script will create some extra TF frames for additional object detections. This can be useful if, say, you don't want the robot to actually see and pick up oranges. Run it very easily with:

rosrun ctp_tom fake_objects.py

If you set the fake:=true option when bringing up planner.launch, the perception system will not start but everything else will. This lets us manually publish extra objects and data.

One-Arm Orange Tests (Old Version)

These are the "old" experiments, and may not work all that well any more.

Downloading Dataset

The first dataset we have consists of a set of demonstrations of TOM picking and moving an "orange" from one place to another. These files are all available on Dropbox:

https://www.dropbox.com/sh/jucd681430959t2/AACGdPQp3z24VineOrYJSK4na?dl=0

Just download them and unpack into whatever location makes sense for you. You'll be running the CTP tool from the directory root after unpacking these data files.

Getting started

The TOM code and scripts are located in the ctp_tom ROS package.

Run the simple TOM simulator and server. This takes in joint positions and will move the arm to those positions:

# No visualization
roslaunch ctp_tom simple_sim.launch

# With visualization
roslaunch ctp_tom simple_sim.launch rviz:=True

Then move into your data directory:

roscd ctp_tom
cd data
rosrun ctp_tom tom_test.py

We should see a trajectory appear, and output like:

0 / 11
1 / 11
2 / 11
3 / 11
4 / 11
5 / 11
6 / 11
7 / 11
8 / 11
9 / 11
10 / 11
11 / 11
=== close ===
1 / 11
1 / 11
2 / 11
3 / 11
4 / 11
5 / 11
6 / 11
7 / 11
8 / 11
9 / 11
10 / 11
11 / 11
=== open ===

Execution

You can then execute with the command:

rosrun costar_task_plan tom_test.py --execute

This will publish trajectory goals to joint_state_cmd.

We expect objects will appear as TF frames.