Art / Fida / Family Archive / Meerkat / Tuba / TellTable / Other Brother / SketchStorm / Autonomous Presence
The autonomous presence project explores what it means to live with robotic like devices that have intentionally been designed to steer away from copying human behaviour and aesthetic. This set of small robotic like devices, called rudiments, investigate human-machine interactions in a variety of different ways. Through careful iterations in their design, the rudiments are intended to provoke curiosity and discussion around the possibility of autonomy in interactive systems. They steer away from a humanoid approach to robot design and explore the potential relations and possibilities for combining robotic devices with appliance like characteristics.
Rudiment 1 (see pictures below), the least sophisticated of the three, is made up of two modules connected via a long, flexible cable. One of the modules quite literally wanders around a magnetic surface, e.g., a fridge door. Its round-shaped wood and acrylic case encapsulates its magnetic wheels and also a narrow range IR sensor to detect nearby movement. Its speed and direction are randomly changed when the IR sensor is triggered. The second module, a switchbox, is magnetically affixed to the same surface as the moving module. It simultaneously provides power and sends signals to the moving module whenever its own wide-range IR sensor detects peripheral movement. On receiving a signal, the moving module is activated and moves for a random amount of time (limited by a set min- and maximum). To prevent it from falling off a surface, two sensors protrude from the front of the moving module. Each sensor contains two switches, one to detect an obstacle and the other to detect when an edge is reached. If these sensors are triggered, the module backs up and changes its direction. Both modules contain Arduino micro-controller boards to control the sensors, actuation and communication functions.
Rudiment 2, made by Alex Taylor, consists of a plywood servomotor and base (see picture below), and two acrylic microphone cases—all three wirelessly connected using the Zigbee standard. The servomotor, with an articulated arm and pencil attached, is slotted into the middle of the wooden base. As well as its mechanical parts, the bespoke servomotor houses a customised Arduino micro-controller and an FIO board1 with XBee module (the latter enabling wireless communication). The two encased microphones function as a trigger for the servomotor and the arm/pencil attachment. The rotation and direction of the motor’s arm are dictated by the level of sound input and which of the two microphones detects a louder sound. Also, using a simple caching method, the system’s sensitivity is varied: sustained or particularly loud noises make it increasingly sensitive and consequently the motor arm’s frequency and degree of rotation were increased. The intended effect is a machine that appears to draw in response to sounds but, to some degree, controls its movements. The rudiment’s output is drawn on removable paper sheets that in effect visually record a soundscape.
Rudiment 3 consists of an acrylic cog system and casement suspended on a horizontally extended, toothed belt of adjustable length (see pictures below). Using suction cups, the flexible belt can be mounted on any smooth vertical surface, e.g., a window, under the cog system, the oval-shaped casement contains a video camera monitoring the environment. Actuated by a DC motor, the cog system moves the casement left and right along the belt’s entire length. The camera can also be rotated left or right by up to 70 degrees. These movements effectively change the viewing direction of the camera. In addition, the opaque casement can display eight different colours using three integrated LEDs (red, green, and blue).
The rudiment’s movement and colour are controlled by an Arduino micro-controller, which in turn communicates with a small PC encased in a wooden box. The PC receives the video signal from the camera and triggers the rudiment’s behaviour through a set of simple yet nondeterministic computer vision processes. The program, written in C++, by Xiang Cao, searches for human faces in the video frame using an object detection algorithm based on the Haar classifier cascade. Each time a face is detected, the rudiment will adjust itself by either moving along the belt or turning the camera (randomly choosing between the two actions), so that the face remains centred. As this happens continuously, the camera appears to follow any movement of a detected face. When more than two faces are detected, the rudiment randomly chooses one face to follow. In addition to faces, the rudiment also responds to gross motion in the camera view, momentarily turning towards it. Finally, the rudiment occasionally makes random movements, adding a degree of ambiguity to its behaviour.
In addition to movement, the rudiment turns one of its eight possible colours when a face is detected. The colour is chosen by comparing the detected face with eight face categories, each consisting of three sample faces. The category that contains the most similar face is chosen and the associated colour displayed. The face comparison is based on a straightforward pixel level comparison, without leveraging any predefined knowledge of facial features. This results in a somewhat “machine-defined” similarity measure, which may or may not appear recognisable to users. With a small probability (0.05), the newly detected face may replace an old face sample in the category. As such, the face categories gradually evolve as the rudiment is exposed to more faces. This simple machine learning technique allows the rudiment to adapt to the people who interact with it and present the same colour each time it recognises similar facial features.
s.jpg)
After our final implementations, the three rudiments were installed in two households (both in the South East region of the UK). Seb and Mon, a couple living in the first household, had the machines for four weeks. The second household, made up of Adam and Ania, again a couple, had them for just under four weeks. Both couples were able to choose where to install the rudiments, although, as we intended, their placement was restricted by their design. Each couple were interviewed for approximately an hour when we removed the rudiments.
Together with Alex and Xiang the results of the deployment were written up in a paper which has been accepted at TEI 2010. In order to see the rudiments in action click here to watch a short movie-clip.
Prototyping/exploratory process
The top right image shows the webcam deliberated from its cover and with an extended power and signal lead in order to create more freedom whenever positioning the object. The two images in the middle below show two different setups of how to attach the object to the string. The left being an earlier version compared to the one on the right that was enabled to carry a servo motor alongside in order to actuate the part containing the camera for face-recognition.
The image on the bottom left shows two little reed-contact switches that enable the object to determine whenever it reaches the end of its movement capabilities either being totally on the left or right side on the belt. The image on the right shows the 'eye-part' of the object that is attached to the servo-motor and contains the camera, RGB LEDs and two vibrating motors. The integration of these actuators allows the object to become very expressive in its behaviour.
The images above show a connector I made in order to give users the freedom to quickly adapt the length of the belt in order to fit a specific surface. Where we initially used tape in order to attach the belt to the suction cups, this simple locking mechanism allows continuous change in setup. The clip can be taken of the metal pin attached to the suction cup. By tilting the lock on the middle the clip can be moved around in order to change the length of the belt. Repositioning the lock and sliding the object into the metal pin, the system gets locked again.
The pictures above show some early tests with regards to its movement, camera rotation and face recognition software.
Magnet robot
Living on the fridge the magnet robot has magnet wheels that keep it attached to the fridge surface. Below is an early exploration in which the magnets are located in the bottom centre of the device (with non magnetic wheels to drive the object). Although the distance of the magnet towards the fridge surface could be adjusted by adjusting the screws that attach it to the body; it turned out to be very difficult to get the right 'force-friction' ratio.
The pictures below show the initial setup of the magnet robot in which I replaced the original wheels with magnets. In addition, the pictures show the two motors, motor drivers and the arduino nano that functions as a base for the object. Using the arduino nano with this setup allows the robot to be very easily re-programmed.
In addition to the object as such I also explored a detection mechanism for detecting both an edge and an obstacle. Deliberating a large variety of different possibilities, of which an optical sensor or simple mechanical antenna would seem most feasible, I decided to create a simple mechanical detection mechanism existing of two identical parts. The picture below shows the parts of one mechanism I iterated towards during the exploration of several different forms and material thicknesses.
The basic functionality of the mechanism is extremely simple; it exists of an upper and lower part that can slide over each other, the upper part is connected to the robot. The lower part is segmented and can bend down in case of an edge. The upper part contains a magnet being pulled towards the magnet surface and thus pressing onto an aluminium plate in the bottom part. As long as the magnet is pressing onto the aluminium plate the circuit is closed and no obstacle or edge is detected (see the first image below on the left). However, when an obstacle hits one of the mechanisms the lower part moves back and a little knob pushes the magnet away from the aluminium plate, breaking the circuit (see the image on the right). In the case of an edge, the segmented lower part will bend down, also breaking the circuit (see second image on the left, in this case I lifted the robot). In order to ensure that an edge or obstacle is detected on time and independent of the angle in which it is approached, the mechanisms are positioned in an angle of 90 degrees. This allows the mechanisms to stick out wider than the magnetic wheels in case it would approach an edge or obstacle almost from the side. When something gets detected, the robot will first move backwards, expecting the route it came from to be clear, after which it will turn in the opposite direction of the triggered mechanism.
