LAB

<latex>{\fontsize{16pt}\selectfont \textbf{Preliminary Measurements for the Employment of Robots in Archaeological Use-wear Analysis}} </latex>

<latex>{\fontsize{12pt}\selectfont \textbf{[Johannes Pfleging]}} </latex>
<latex>{\fontsize{10pt}\selectfont \textit{[Master Project ME]}} </latex>

<latex> {\fontsize{12pt}\selectfont \textbf{Abstract} </latex>

Use-wear analysis is a field in archaeology which investigates human tool use in prehistory by means of traces on lithic tools. Comparing pattern of microscopic traces on replicated stone tools to those on artefacts allows for hypothesis about stone-tool functions. During the experiments the replicas are used according to the task, which is investigated under laboratory conditions. Frequently investigated tasks include the manufacturing of products with hand-held tools. In order to derive linkage between tool use and wear pattern all relevant parameters of the task must be known. Unfortunately, precise monitoring of use-wear relevant parameters during the experiments is rare for these tasks. In this work we designed an experimental setup to measure the dynamics of two prototypical tasks, scraping and whittling.

A strain gage force sensor (ATI) combined with visual tag tracking system (AprilTags) was used to record force, torque as well as position and orientation of hafted flint stone tools. The set-up allows to vary between two tool configurations, distal and perpendicular hafting of flakes of flint stone. We recorded a set of exploratory experiments on sheep hide with varying task parameters for testing of the experimental set-up.

The method shows how a feasible measuring configuration can be set-up for precise monitoring of use-wear experiments. Extension of the set-up to similar tasks related to shafted tools like sawing or hacking is imaginable.

<latex> {\fontsize{12pt}\selectfont \textbf{Backround}} </latex>

Archaeology studies the evolution of the human beeing from our earliest ancestor to the human of today. These studies include the evolution of human anatomy, motoric and cognitive capabilities, cultural and social behavior, etc. Analysis of human remainings is the major source of information about prehistoric life. These include their bodies but also objects that they created and used for their purposes, like tools, weapons, jewellery or art. Because of their non-biological nature, the most durable remainings among these are lithic technologies.

The big field of lithic use-wear analysis can be subdivided into two major fields. The first makes use of experiments to infere wear traces to tool funtion. The second makes use of these findings to identify wear on artefacts which in return gives knowledge about how the related population used the tool. Unfortunately, the most reliable statement, which could be derived with this method so far, concerns the material that has been worked with the tool, e.g. hide, meat, stone or wood. The complete tool motion which is related to a particular tasks could not be identified yet. Reasons for that are believed to be found in non-standardized and non-replicable procederes. In particular, almoast all experiments lack of quantitative data about tool dynamics i.e. pose and force/ torque even though these parameters are known to affect the process of wear formation.

This work presents a measurement configuration which enables to measure with high temporal resolution tool position and orientation as well as applied force and torque on the tool during use. From the position/ orientation one can derive the parts of the tool that were in contact with the worked material during use and corresponding data traces of force/ torque give the ammount of pressure that was applied. With these parameters at hand it is possible to test hypotheses about the relationship between wear pattern and certain dynamic movements.

<latex> {\fontsize{12pt}\selectfont \textbf{Methodology}} </latex>

The central part of the measurement system is the tool (see below). It is derived from replicas and comprises the force sensor and the AprilTag which are linked to peripheral systems for data recording.

The main parts are shaft, force sensor, quadratic tag and tool tips. The wooden shaft is connected to the plane side of the sensor via an adapter made of ABS plastic and screws. The leading side of the sensor is connected to the leading adapter which serves as holder for the tool tips. The tool tips fit into the mounting of the adapter and are locked in place by screws. Additionally a squared tag of 58mm side length is fixed laterally to the adapter with the plane parallel to the long axis of the shaft. The tag is part of the AprilTags tracking system (\cite{Olson2011}). It was printed and placed on a wooden board. The choice of size is a trade-off between accuracy of the tracking system and limitation of tool handling. A soft protection film is wrapped around the sensor and fastened by plastic straps.

The force sensor (ATI Mini45) is connected through a cable to the Interface Board/ Power Supply which in turn is connected to the data acquisition (DAQ) card (National Instruments PCI-6220 M) of the PC. Proprietary software of ATI converts uncalibrated signals of the sensor to calibrated values, which can be saved in text files on the PC. The sensor measures all three components of force and torque which are applied at the leading side of the sensor. It is calibrated to the largest range available: +-580N in Fx and Fy, +-1160N in Fz, +- 20Nm of torque on all axis. Accuracies are less than 11.6 N for force and less than 0.35 Nm for torque.

The position sensor consists of mid-price digital video camera (Panasonic Lumix DMC-FZ200) and the tag on the tool. Videos of the tool are recorded during the experiments. It is important that the printed side of the tag always be in field of view of the camera. At post processing the videos were analyzed using a C++ implementation (http://people.csail.mit.edu/kaess/apriltags/) of AprilTags. This software identifies the tag on the images and determines on the basis of tag size and camera intrinsics the position and orientation relative to the camera. The information comprised in the 2D code on the tag is irrelevant for our application. We used the camera at resolution of 1280×720 pixels and at nominal focal length of 25mm. The camera has rolling shutter technique and it was calibrated using a camera calibration toolbox for MATLAB (http://www.vision.caltech.edu/bouguetj/calib_doc/). For both, the camera and DAQ card we used sample frequencies of 100 Hz.

The tool tips were derived from ethnographically and archaeologically plausible replicas of tools. Splinters of flint stone (flakes) were knapped by experienced archaeologists into so called scrapers, which then were hafted by birch tar either distally along the axis of the haft or perpendicular to it. The shaft was cut and rasped in order to fit into the mounting of the adapter.

Experiments with two different hide-processing tasks were carried out, which are called scraping and whittling. The perpendicularly hafted tool was used for scraping tasks and the distally-hafted tool for whittling. Tool trajectories for both tasks are defined according the the figure below, left. The part of the trajectory when the tool is in contact with the workpiece is called stroke phase and the remaining part when the tool has no contact to the workpiece is called flying phase. Although the tasks can also be performed with non-hafted tools, here hafted tools were used due to the constrains of the measuremnt configuration. Fresh hide as well as dried (6 days under fume hood) hide was used as workpiece. Further variation was introduced by different subjects and variation between dominant and non-dominant hand. Further the individual interpretation of the task description introduces variation in the tool dynamics which is measured by the sensors. The experiments were performed in the lab as illustrated on the figure to the right.

<latex> {\fontsize{12pt}\selectfont \textbf{Result}} </latex>

The figures above shows the data sets of the dynamic parameters over time for a representative cycle of the whittling (left) and scraping task (right). Corresponding frame of reference is defined in the figure to the right. For the whittling task, two tool cycles can be identified, each characterized by two successive minima of the y-position. During elevated values of the z position up to -0.2 m the tool is in the flying phase (i.e. not touching the hide). This can also be seen by zero values for force and torque. The tool is swiftly returned to the starting point of the next stroke phase, which can be seen in the steep rise of y position. The stroke phase can be identified by constant z position and non-zero values for force and torque. The y position decreases slowly from maximum to the following minimum. Within one cycle orientation of the tool varies mainly in yaw angle which is equivalent to the working angle. Furthermore, the slight variation in x position indicates that tool trajectory is not limited to one y-z plane. The starting point of the stroke phase in x direction of the first cycle is shifted about 0.02m compared to the starting point of the second cycle.

<latex> {\fontsize{12pt}\selectfont \textbf{Discussion}} </latex>

The experimental set-up allows us to measure the force and the torque applied between stone tip and shaft as well as position and orientation of the tool over time. The measurement data was processed to illustrate the tool motion and corresponding force/ torque in different reference frames, i.e. in different perspectives. The next step would be to analyse the wear pattern that was created during the experiments and to correlate the dynamic parameters against it. The optical analysis of wear traces requires microscopes and a bunch of statistical image processing techniques. Beside that the experimental study has to be designed accordingly. That means that each experiment has to performed with one tool for a sufficient number of cycles and established cleaning proceders of the tool have to be applied before the optical analysis.

In order for the set-up to be suitable for the creation of accurate, large sets of experiments the following should be improved:

• Calibration of the camera axis: The working angle can only be derived if the orientation of the workpiece relative to the camera is known. Therefore, the table and the camera have to be oriented horizontally. So far this has been done before each experiment seession using a level. This process could be facilitated using some stationary apparatus.
  

• Syncronization of the data traces: So far the data traces of force and position were synconized manually by identification of an artificially evoked incidence in the time plots. This procedure introduces inaccuracies depending on the sampling time.
  

• Measurement of the tool morphology: The dimensions of the tool and the lithic tips were measured using the caliber and photogrammetry. Both techniques are time consuming. Faster methods like 3D laser scan are beneficial.
  

• Limitations to hafted tools: Inherently the measurement configuration only allows the use of hafted tools. Other configurations must be sought to enhance the set-up for non-hafted tools.
  

• Limitations of AprilTags: The tracking software only tracks the tag for velocities up to 0.9m/sec. Above that the motion blur in the images is to high. A solution might be the use of global shutter camera techniques. The data traces of the orientation contains artefacts such as steps, mostly for roll and yaw angle at fast motions fo the tag. A more robust technique would be the integration of a calman filter into the AprilTags software.
  

<latex> {\fontsize{12pt}\selectfont \textbf{Downloads}} </latex>