Data format

Data files are generally named with a .dat extension.

Each data file contains a collection of frames (snapshots) of sensor data. The file is organized as a collection of lines. Each line begins with a keyword, followed by some information. Any line beginning with a word other than a known keyword is considered to be a comment line. In a file there are required keywords and (possibly) optional keywords. The required keywords are @OPEN, @SENS, #ROBOT.

@OPEN date time

Example: @OPEN 10-12-93 16:03:00.557068

This marks the first entry in a sequence, stating when that sequence starts. A sequence is simply a sequence of robot data snapshots and (optionally) actions. Date = mm-dd-yy, Time = hr:mm:ss.microsecs.

@SENS date time

Example: @SENS 10-07-93 20:07:43.800980

Marks the beginning of a new sensation, or snapshot of sensor data.

#ROBOT x y theta

Example: #ROBOT 0.000148 2.256479 0.007522

x,y and theta give the sensed (via dead reckoning) robot position. At the start of each sequence, x, y theta are all 0. x and y are in cm's, theta is in degrees.

#SONAR n: s1 s2 .. s24

Example: #SONAR 24: 807.239990 249.455994 148.871994 121.440002 103.152000 ...

n is the number of sonar readings s1..s24 sonar readings ranging from 8 cm to 808 cm. No sonar echo is reported as a maximum sonar range reading.

#LASER n: x1 y1 x2 y2 ... xn yn

Example: #LASER 48: -14.114452 352.500000 0 0 -7.618582 352.500000 ...

n is the number of laser readings xi and yi are the x and y coordinates of the ith laser reading, in a coordinate frame centered at the direct center of the robot, with the y axis (x=0) pointing straight forward. y varies from 40 cm to 350 cm. x varies No laser return is reported as coordinate (0 0).

#LASER-Y n: y1 y2 ..yn

Example: #LASER-Y 51: ...

n number of laser readings (only y-coordinates) y1..yn laser readings varing from 40 cm to 350 cm. No laser return is reported as a zero reading

#CAMERA tilt pan n: l11 l12 l13 l21 l22 l23 ... (groups of three)

Example: #CAMERA -23.0 12.2 3: 12 344 1200

tilt = tilt angle (in deg)

pan = pan angle (in deg)

n : number of markers found * 3, ranges from 0 to 6 (all different color) (n is exactly the number of numbers that follows, and there are 3 numbers per marker. Divide this value by three to get the number of markers found)

lj1: horizontal pixel value (x) in camera coordinates

lj2: vertical pixel value (y) in camera coordinates

lj3: strength of the detected marker (by find.c)

#VISION x y xscale yscale:rgb*(x*y)

Example: #VISION 150 124 1.000000 1.000000:...

x and y give the dimensions of the image. xscale and yscale give the scaling factors for x and y (not used thus far). This is followed by a colon, followed by the x*y color pixel values. Each pixel is described by three values: the red, green, and blue intensity levels. Each of these is stored as a single 8-bit integer (which reads out as a single ASCII character in most editors). The original camera image is 600x496, and is converted into a 150x124 image by reducing squares of 4 pixels into a single pixel (averaging values). The (0,0) coordinate of the image is the top left corner; (150,124) is bottom right.

#RECTANGLE name x1 y1 x2 y2 x-res y-res: r1 g1 b1 ... rn gn bn where n= {(x2-x1)/xres}*{(y2-y1)/y-res}

Example: #RECTANGLE biggie 0 0 150 124 75 124: 59 87 250 84 23 190

Describes a set of r,g,b intensity values of rectangles from the #VISION image. For instance, the above example defines the feature "biggie" as two 75x124 rectangles within a single 150x 124 rectangle begining at (0 0). The three values are the average r,b,and g values over the two rectangles. These features can be generated automatically by a program. See the file example.dat and sample.conf for more examples.

@ACTION datestamp timestamp

Example: #ACTION 10-07-93 20:07:43.800980 POTENTIAL-FIELD-GOTO 150 0

Describes an action performed by the robot, at a specific time (when the action is initiated). These may be interleaved with sensations.

#ROBOT x y orientation in_motion?

Example: #ROBOT 500 500 90 0

x, y, orientation are the dead-reconing coordinates of the robot in_motion? flag: 1, if robot was in motion at this time, 0 if not

#SET-ROBOT x y orientation

Example: #SET-ROBOT 500 500 90

Command that sets the internal dead-reconing coordinates for the base. x y orientation are the new coordinates

#SPEED forward-speed rotation-speed

Sets the max-speed values for the robot base. forward-speed rotation-speed are in cm/sec and deg/sec

#SET-GOAL action_nr dist angle x_target y_target

Example: #SET-GOAL 3 125 0 514.147 653.179 Issues a set-goal command. This command navigates the robot to a specified position using the potential-field obstacle avoidance mechanism. and specify the goal point in polar coordinates relative to the robot, and and specify the same target point in the global world coordinate system. The latter two parameters are the actual parameters in the TCA-SET-GOAL command. refers to a list of 7 actions used by Sebastian for his learning stuff, and will be irrelevant for anybody else. If you look through his data files you will find that only 7 different actions are invoked at all.

#STOP

The robot received a STOP command.

#CATEGORY label:

Describes the hand-input categorization of a sensation. At present time, the value of 1 indicates a scene containing a door, 2 indicates a scene containing a bin, 0 indicates a scene containing neither. No scene contains both a door and a bin simultaneously.

Overall Points:

Keywords beginning with @ describe the format of sequences (such as the time data collection commenced or the time a given sensation was sampled).

Keywords beginning with # describe sensations and actions of the robot.

Last Updated: 9July94 12:00 josullvn+@cs.cmu.edu