Only released in EOL distros:  

cram_core: cram_designators | cram_execution_trace | cram_language | cram_math | cram_process_modules | cram_projection | cram_reasoning | cram_test_utilities | cram_utilities

Package Summary

A common interface definition for lower level control routines.

Contents

  1. Introduction
  2. API

Introduction

One goal of CRAM is to execute the same high-level plan on various different robots. For that, we need to find an abstraction to separate robot specific components and the high-level parts that specify the actual behavior. For that purpose, we introduce process modules. Process modules are basically the parts of a CRAM system that directly communicate with the outer world, i.e. that trigger the navigation process (e.g. move_base), manipulation or perception. For the high-level components of the system, these process modules must be black boxes in order to keep them general. The process modules for controlling a robot are started up at the beginning of plan execution and shut down when the robot finished its actions. Their input is always an action designator provided by the cram_designators package.

API

Process modules are defined with the macro def-process-module. The definition looks similar to a function definition in Common Lisp with the difference that only one parameter can be declared, the name of the variable the input designator is bound to. Whenever input is received, the body of the process module is executed. Basically, action execution in process modules has to perform the following steps:

  1. Resolve the action designator to get the actual commands to send to the robot's hardware.
  2. Execute these parameters.
  3. Either throw an error if the execution failed or optionally return a result value that is passed to the caller.

The following code illustrates how a simple navigation process module could be defined:

(def-process-module navigation (input-designator)
  (let ((goal-pose (reference input-designator)))
    (or (execute-navigation-action goal-pose)
        (fail 'navigation-failed))))

We assume that the input designator (an action designator) resolves to a 3d pose that we can send to the function execute-navigation-action. The function returns NIL if navigation failed. In that case we throw an error.

To send a command to a process module, the package provides the API method pm-execute. The following example shows how to send a navigation action to the process module defined above:

(top-level
  (with-process-modules-running (navigation)
    (pm-execute navigation (make-designator 'action `((type navigation) (goal ,some-pose)))))

We assume that the variable some-pose is already bound to a 3d pose. The macro with-process-modules-running is used to start up the specified process modules and to shut them down after the code finished. Normally, the macro is used at top-level and the process modules keep running over the complete runtime of the robot. The surrounding top-level form is necessary because with-process-modules-running needs to be executed in a cram_language environment.

Wiki: cram_process_modules (last edited 2012-09-19 15:47:24 by Lorenz Mösenlechner)