PostureEngine.h

Go to the documentation of this file.

//-*-c++-*-
#ifndef INCLUDED_PostureEngine_h
#define INCLUDED_PostureEngine_h

#include "Motion/OutputCmd.h"
#include "Motion/Kinematics.h"
#include "Shared/RobotInfo.h"
#include "Shared/LoadSave.h"

class WorldState;

//! A class for storing a set of positions and weights for all the outputs
/*! File Format: ([..] indicates an optional parameter)\n
 *  - First line: '<tt>\#POS</tt>'
 *  - Followed by a series of:
 *    - '<tt>specialize </tt><i>[regex]</i>' - only robots whose model name matches the regular
 *      expression will attempt to parse lines between this and the next 'specialize' command.  An empty
 *      (absent) regex matches all robots (equivalent to '<tt>specialize&nbsp;.*</tt>')
 *    - '<tt>condensed </tt><i>model</i>' - specifies model name for which condensed specifications will be
 *      interpreted
 *    - '<tt>verbose</tt>' - switches the load/save format back to "verbose" style, where each output/sensor
 *      value is listed individually
 *    - '<tt>&lt;</tt><i>section</i><tt>&gt;</tt> ... <tt>&lt;/</tt><i>section</i><tt>&gt;</tt>' - specifies a
 *      section for following "verbose" style lines (valid <i>section</i> values listed below)
 *    - '<i>section</i> <i>value1</i> <i>value2</i> <i>value3</i> ...' - specify all values for a section in
 *      one line, "condensed" style (valid <i>section</i> values listed below).  Must have a value for every
 *      item in the section (no extra or missing values)
 *    - '<i>output-name</i> <i>value</i> <i>[weight]</i>' - specifies an output value, with optional weighting
 *      value (if no weight specified, '1' is assumed)
 *  - Last line: '<tt>\#END</tt>'
 *  
 *  Note that '=' can be used to separate tokens as well as any whitespace character.
 *  All angle values should be specified in radians.
 *
 *  Valid section names are:
 *  - <tt>meta-info</tt> - only recognizes two fields: <tt>timestamp</tt> and <tt>framenumber</tt>
 *  - <tt>outputs</tt> - supports all output names (e.g. ERS7Info::outputNames)
 *  - <tt>buttons</tt> - supports all button names (e.g. ERS7Info::buttonNames)
 *  - <tt>sensors</tt> - supports all sensor names (e.g. ERS7Info::sensorNames)
 *  - <tt>pidduties</tt> - supports output names corresponding to PID joints (this is the "duty cycle" for the joint)
 *
 *  Additionally 'weights' can be used as a condensed section name (but not a
 *  verbose tag-style section, since weights are specified on each line).  The
 *  'weights' specification must follow the 'outputs' specification to be effective.
 *
 *  Data following '\#' is ignored as comments.  Be aware if you
 *  load the file and then save it again, these comments will be lost.
 *
 *  Example 1: Specifies only neck joints, looks forward and up a little
<table><tr><td align=left><pre>
\#POS 
NECK:tilt 0
NECK:pan  0
NECK:nod  0.2
\#END
</pre></td></tr></table>
 *  All other, unspecified, joints will be set to weight=0.
 *
 *  Example 2: Logged sensor data from an ERS-7 robot
<table><tr><td align=left><tt>
\#POS<br>
condensed ERS-7<br>
meta-info = 1439041 178803<br>
outputs = -0.131722 0.148077 1.74592 -0.30276 0.341717 2.13361 -1.03935 0.091124 2.18958 -0.804097 0.034171 1.67458 -0.016362 -0.089143 0.125563 0.407243 -0.054399 -0.064704 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0<br>
buttons = 0 0 0 0 0 0 0 0 0 1<br>
sensors = 500 860.139 425 -2.06456 -0.516139 -8.94642 0.99 30.63 2200 8.265 0<br>
pidduties = -0.164062 0.0878906 0.0957031 0.246094 -0.0195312 -0.0546875 -0.164062 -0.0195312 0.164062 0.0273438 0 -0.0683594 -0.00585938 -0.00390625 0.0820312 -0.0390625 0.015625 0.00390625<br>
\#END<br>
</tt></td></tr></table>
 *  Condensed posture files can still be loaded on other models.  For each entry, RobotInfo::Capabilities will be
 *  used to map from the model specified by the <tt>condensed</tt> command to the current host.
 *
 *  @see PostureMC
 *  @see <a href="http://www.cs.cmu.edu/~tekkotsu/Kinematics.html">Tekkotsu's Kinematics page</a>
 *  @see <a href="http://www.cs.cmu.edu/~dst/Tekkotsu/Tutorial/postures.shtml">David Touretzky's "Postures and Motion Sequences" Chapter</a>
 *  @see <a href="http://www.cs.cmu.edu/~dst/Tekkotsu/Tutorial/forwardkin.shtml">David Touretzky's "Forward Kinematics" Chapter</a>
 *  @see <a href="http://www.cs.cmu.edu/afs/cs/academic/class/15494-s06/www/lectures/postures.pdf">CMU's Cognitive Robotics posture slides</a>
 *  @see <a href="http://www.cs.cmu.edu/afs/cs/academic/class/15494-s06/www/lectures/kinematics.pdf">CMU's Cognitive Robotics kinematics slides</a>
 */
class PostureEngine : public LoadSave, public Kinematics {
public:

  //!@name Constructors
  
  //!constructor
  PostureEngine() : LoadSave(), Kinematics(*kine), saveFormatCondensed(false), saveSensors(NULL), loadSensors(NULL) {}
  //!constructor, loads a position from a file
  /*! @todo might want to make a library stored in memory of common positions so they don't have to be loaded repeatedly from memstick */
  PostureEngine(const std::string& filename) : LoadSave(), Kinematics(*kine), saveFormatCondensed(false), saveSensors(NULL), loadSensors(NULL) { loadFile(filename.c_str()); }
  //!constructor, initializes joint positions to the current state of the outputs as defined by @a state
  PostureEngine(const WorldState* st) : LoadSave(), Kinematics(*kine), saveFormatCondensed(false), saveSensors(NULL), loadSensors(NULL) { if(st!=NULL) takeSnapshot(*st); }

  //! copy constructor
  PostureEngine(const PostureEngine& pe)
    : LoadSave(pe), Kinematics(pe), saveFormatCondensed(pe.saveFormatCondensed), saveSensors(pe.saveSensors), loadSensors(NULL)
  {
    for(unsigned int i=0; i<NumOutputs; i++)
      cmds[i]=pe.cmds[i];
  }
    
  //! assignment operator
  PostureEngine& operator=(const PostureEngine& pe) {
    LoadSave::operator=(pe);
    Kinematics::operator=(pe);
    saveFormatCondensed=pe.saveFormatCondensed;
    saveSensors=pe.saveSensors;
    for(unsigned int i=0; i<NumOutputs; i++)
      cmds[i]=pe.cmds[i];
    return *this;
  }

  //! destructor
  virtual ~PostureEngine();
  //@}



  //! You should be able to call the non-virtual functions without checking out, just a MotionManager::peekMotion().  Theoretically.
  //!@name Output Value Access/Control
  virtual void takeSnapshot(); //!< sets the values of #cmds to the current state of the outputs (doesn't change the weights)
  virtual void takeSnapshot(const WorldState& st); //!< sets the values of #cmds to the current state of the outputs as defined by @a state (doesn't change the weights)
  virtual void setWeights(float w) { setWeights(w,0,NumOutputs); } //!< set the weights of all #cmds
  virtual void setWeights(float w, unsigned int lowjoint, unsigned int highjoint); //!< the the weights of a range of #cmds
  virtual void clear(); //!< sets all joints to unused
  inline PostureEngine& setOutputCmd(unsigned int i, const OutputCmd& c) { cmds[i]=c; return *this; } //!<sets output @a i to OutputCmd @a c, returns @c *this so you can chain them; also remember that OutputCmd support implicit conversion from floats (so you can just pass a float)
  inline OutputCmd& operator()(unsigned int i) { return cmds[i]; } //!< returns output @a i, returns a reference so you can also set through an assignment to this call, e.g. pose(MouthOffset)=.1; (remember that OutputCmd support implicit conversion from floats)
  inline const OutputCmd& operator()(unsigned int i) const { return cmds[i]; } //!< returns output @a i
  inline OutputCmd& getOutputCmd(unsigned int i) { return cmds[i]; } //!< returns output @a i, returns a reference so you can also set through an assignment
  inline const OutputCmd& getOutputCmd(unsigned int i) const { return cmds[i]; } //!< returns output @a i
  //@}



  //!Uses LoadSave interface so you can load/save to files, uses a human-readable storage format
  //!@name LoadSave
  virtual void setSaveFormat(bool condensed, WorldState* ws); //!< sets #saveFormatCondensed and #loadSaveSensors (pass ::state for @a ws if you want to use current sensor values)
  virtual unsigned int getBinSize() const;
  virtual unsigned int loadBuffer(const char buf[], unsigned int len, const char* filename=NULL);
  virtual unsigned int saveBuffer(char buf[], unsigned int len) const;
  virtual unsigned int loadFile(const char filename[]);
  virtual unsigned int saveFile(const char filename[]) const;
  
  //! used for storing sensor data loaded from file
  struct SensorInfo {
    SensorInfo() : buttons(), sensors(), pidduties(), timestamp(-1U), frameNumber(-1U) {}
    std::map<unsigned int,float> buttons; //!< button values
    std::map<unsigned int,float> sensors; //!< sensor values
    std::map<unsigned int,float> pidduties; //!< pid duty cycle values, indices are relative to PIDJointOffset
    unsigned int timestamp; //!< timestamp associated with sensor values, or -1U if none found
    unsigned int frameNumber; //!< sensor frame number associated with sensor values, or -1U if none found
  };
  virtual void setLoadedSensors(SensorInfo* si) { loadSensors=si; } //!< if @a ws is non-NULL, any sensor values in loaded postures will be stored there (otherwise they are ignored)
  virtual SensorInfo* getLoadedSensors() const { return loadSensors; } //!< returns value previously stored by setLoadSensors()
  //@}



  //!@name Kinematics

  //! Performs inverse kinematics to solve for positioning @a Peff on link @a j as close as possible to @a Ptgt (base coordinates in homogenous form); if solution found, stores result in this posture and returns true
  /*! @param Ptgt the target point, in base coordinates
   *  @param link the output offset of the joint to move
   *  @param Peff the point (relative to @a link) which you desire to have moved to @a Ptgt (it's a point on the effector, e.g. (0,0,0) if you want to move the effector's origin to the target)
   *
   *  The difference between solveLinkPosition() and solveLinkVector() is typically small,
   *  but critical when you're trying to look at something -- the solution obtained by
   *  simplying trying to solve for the position may not align the vector with the target --
   *  solveLinkVector() tries to ensure the vector is aligned with the target, even if that
   *  isn't the closest solution position-wise.
   */
  virtual bool solveLinkPosition(const fmat::SubVector<3,const float>& Ptgt, unsigned int link, const fmat::SubVector<3,const float>& Peff);

  //! Performs inverse kinematics to solve for positioning Peff on link @a j as close as possible to @a Ptgt (base coordinates); if solution found, stores result in this posture and returns true
  /*! @param Ptgt_x the target x position (relative to base frame)
   *  @param Ptgt_y the target y position (relative to base frame)
   *  @param Ptgt_z the target z position (relative to base frame)
   *  @param link the output offset of the joint to move
   *  @param Peff_x the x position (relative to @a link) which you desire to have moved to @a Ptgt (it's a point on the effector)
   *  @param Peff_y the y position (relative to @a link) which you desire to have moved to @a Ptgt (it's a point on the effector)
   *  @param Peff_z the z position (relative to @a link) which you desire to have moved to @a Ptgt (it's a point on the effector)
   *
   *  The difference between solveLinkPosition() and solveLinkVector() is typically small,
   *  but critical when you're trying to look at something -- the solution obtained by
   *  simplying trying to solve for the position may not align the vector with the target --
   *  solveLinkVector() tries to ensure the vector is aligned with the target, even if that
   *  isn't the closest solution position-wise.
   */
  virtual bool solveLinkPosition(float Ptgt_x, float Ptgt_y, float Ptgt_z, unsigned int link, float Peff_x, float Peff_y, float Peff_z)
  { return solveLinkPosition(fmat::pack(Ptgt_x,Ptgt_y,Ptgt_z),link,fmat::pack(Peff_x,Peff_y,Peff_z)); }

  //! Performs inverse kinematics to solve for aligning the vector through Peff on link @a j and the link's origin to point at @a Ptgt (base coordinates in homogenous form); if solution found, stores result in this posture and returns true
  /*! @param Ptgt the target point, in base coordinates
   *  @param link the output offset of the joint to move
   *  @param Peff the point (relative to @a link) which you desire to have moved to @a Ptgt (it's the desired "effector")
   *
   *  The difference between solveLinkPosition() and solveLinkVector() is typically small,
   *  but critical when you're trying to look at something -- the solution obtained by
   *  simplying trying to solve for the position may not align the vector with the target --
   *  solveLinkVector() tries to ensure the vector is aligned with the target, even if that
   *  isn't the closest solution position-wise.
   */
  virtual bool solveLinkVector(const fmat::SubVector<3,const float>& Ptgt, unsigned int link, const fmat::SubVector<3,const float>& Peff);

  //! Performs inverse kinematics to solve for aligning the vector through Peff on link @a j and the link's origin to point at @a Ptgt (base coordinates); if solution found, stores result in this posture and returns true
  /*! @param Ptgt_x the target x position (relative to base frame)
   *  @param Ptgt_y the target y position (relative to base frame)
   *  @param Ptgt_z the target z position (relative to base frame)
   *  @param link the output offset of the joint to move
   *  @param Peff_x the x position (relative to @a link) which you desire to have moved to @a Ptgt (it's the desired "effector")
   *  @param Peff_y the y position (relative to @a link) which you desire to have moved to @a Ptgt (it's the desired "effector")
   *  @param Peff_z the z position (relative to @a link) which you desire to have moved to @a Ptgt (it's the desired "effector")
   *
   *  @todo this method is an approximation, could be more precise, and perhaps faster, although this is pretty good.
   *
   *  The difference between solveLinkPosition() and solveLinkVector() is typically small,
   *  but critical when you're trying to look at something -- the solution obtained by
   *  simplying trying to solve for the position may not align the vector with the target --
   *  solveLinkVector() tries to ensure the vector is aligned with the target, even if that
   *  isn't the closest solution position-wise.
   */
  virtual bool solveLinkVector(float Ptgt_x, float Ptgt_y, float Ptgt_z, unsigned int link, float Peff_x, float Peff_y, float Peff_z)
  { return solveLinkVector(pack(Ptgt_x,Ptgt_y,Ptgt_z),link,pack(Peff_x,Peff_y,Peff_z)); }
  
  //! Performs inverse kinematics to solve for positioning @a Peff on link @a j as close as possible to @a Ptgt (base coordinates in homogenous form); if solution found, stores result in this posture and returns true
  /*! @param oriTgt the target orientation, as a quaternion relative to base frame
   *  @param link the output offset of the joint to move
   *  @param oriOffset an orientation offset (relative to @a link) which you desire to have moved to @a oriTgt
   *
   *  The quaternion functions of fmat are helpful for creating axis-aligned rotations: fmat::quatX(r), fmat::quatY(r), fmat::quatZ(r), and fmat::QUAT_IDENTITY
   */
  virtual bool solveLinkOrientation(const fmat::Quaternion& oriTgt, unsigned int link, const fmat::Quaternion& oriOffset);
  
  //! Performs inverse kinematics to solve for both a position and orientation, bringing @a posOffset on @a link to @a posTgt, in the orientation specified.
  /*! @param posTgt the target point, in base coordinates
   *  @param oriTgt the target orientation, as a quaternion relative to base frame
   *  @param link the output offset of the joint to move
   *  @param posOffset the point (relative to @a link) which you desire to have moved to @a posTgt (it's the desired "effector")
   *  @param oriOffset an orientation offset (relative to @a link) which you desire to have moved to @a oriTgt
   *
   *  The quaternion functions of fmat are helpful for creating axis-aligned rotations: fmat::quatX(r), fmat::quatY(r), fmat::quatZ(r), and fmat::QUAT_IDENTITY
   */
   virtual bool solveLink(const fmat::SubVector<3,const float>& posTgt, const fmat::Quaternion& oriTgt, unsigned int link, const fmat::SubVector<3,const float>& posOffset, const fmat::Quaternion& oriOffset);

  //@}



  //!@name Combining Postures
  
  //! sets joints of this to all joints of @a pe which are not equal to unused (layers @a pe over this) stores into this
  virtual PostureEngine& setOverlay(const PostureEngine& pe);
  //! sets joints of this to all joints of @a pe which are not equal to unused (layers @a pe over this) returns new PostureEngine
  virtual PostureEngine createOverlay(const PostureEngine& pe) const;

  //! sets joints of this which are equal to unused to @a pe, (layers this over @a pe) stores into this
  virtual PostureEngine& setUnderlay(const PostureEngine& pe);
  //! sets joints of this which are equal to unused to @a pe, (layers this over @a pe) returns new PostureEngine
  virtual PostureEngine createUnderlay(const PostureEngine& pe) const;

  //! computes a weighted average of this vs. @a pe, @a w being the weight towards @a pe (so @a w==1 just copies @a pe)
  virtual PostureEngine& setAverage(const PostureEngine& pe,float w=0.5);
  //! computes a weighted average of this vs. @a pe, @a w being the weight towards @a pe (so @a w==1 just copies @a pe)
  virtual PostureEngine createAverage(const PostureEngine& pe,float w=0.5) const;

  //! computes a weighted average of this vs. @a pe, using the weight values of the joints, storing the total weight in the result's weight value
  virtual PostureEngine& setCombine(const PostureEngine& pe);
  //! computes a weighted average of this vs. @a pe, using the weight values of the joints, storing the total weight in the result's weight value
  virtual PostureEngine createCombine(const PostureEngine& pe) const;

  //! returns the sum squared error between this and pe's output values, but only between outputs which are both not unused
  /*! @todo create a version which does weighted summing?  This treats weights as all or nothing */
  virtual float diff(const PostureEngine& pe) const;
  
  //! returns the average sum squared error between this and pe's output values for outputs which are both not unused
  /*! @todo create a version which does weighted summing?  This treats weights as all or nothing */
  virtual float avgdiff(const PostureEngine& pe) const;
  
  //! returns the max error between this and pe's output values for outputs which are both not unused
  /*! @todo create a version which does weighted summing?  This treats weights as all or nothing */
  virtual float maxdiff(const PostureEngine& pe) const;
  
  //@}

protected:
  //! enumeration of the different section types that may be used as section tags in verbose mode
  enum section_t {
    SECTION_METAINFO, //!< includes timestamp and framenumber
    SECTION_OUTPUTS, //!< entries corresponding to NumOutputs
    SECTION_BUTTONS, //!< entries corresponding to NumButtons
    SECTION_SENSORS, //!< entries corresponding to NumSensors
    SECTION_PIDDUTIES //!< entries corresponding to NumPIDJoints
  };
  //! helper function for loadBuffer, called for each individual line
  virtual bool loadLine(unsigned int linenum, const char* filename, const std::map<std::string,section_t>& sectionMap, std::vector<std::string>& words, section_t& curSection, const Capabilities*& caps, bool& filtered);
  //! helper function for loadLine, strips trailing '~'s from output names to provide backward compatability (note pass by reference, operates 'in-place' on @a word)
  /*! (originally, all output names were uniform length and used '~'s as padding... ugh.) */
  void stripTildes(std::string& str) {
    str.erase(str.find_last_not_of('~')+1); // if n is npos, npos is -1U, so becomes 0... should be safe :)
  }

  //! overriding Kinematics::update, make all updates come from this posture engine's own state, not WorldState
  virtual void update() const;

  //!the table of outputs' values and weights, can be accessed through setOutputCmd() and getOutputCmd()
  OutputCmd cmds[NumOutputs];

  bool saveFormatCondensed;      //!< requests a condensed file format, smaller but less readable
  WorldState* saveSensors;   //!< If non-null, saves will include sensor readings from here
  SensorInfo* loadSensors; //!< if non-null, loads will store any included sensor information here
};

/*! @file
 * @brief Describes PostureEngine, a base class for managing the values and weights of all the outputs
 * @author ejt (Creator)
 */

#endif