In the early 1970's the Freddy robot was built at Edinburgh by Stephen Salter,
with the electronic interface by Harry Barrow, later assisted by Gregan
Crawford. A number of us, including Pat Ambler Harry Barrow (now at Aberdeen
University), Rod Burstall, Chris Brown (now at Rochester) and myself set out
to implement a demonstration of the feasibility of robotic assembly using
vision and touch. The paradigm employed was that a kit of parts for a (wooden)
toy car was dumped from a box into Freddy's workspace and the robot was
required to assemble them. Later demonstrations included both car and ship
assembly.
We employed Freddy's video cameras to acquire a digitised image of the parts, as they rested on the table. Figure ground distinction was made possible by employing light-coloured parts on a dark background. An oblique, wide angle camera was used to locate where parts were to be found, possibly heaped together. A narrower angle camera, looking vertically, was then used to examine each cluster of parts. If Freddy recognised any given part, he moved it to a reserved area of his workspace. Unrecognised objects were treated as "heaps", and Freddy employed one of three strategies to separate parts from a heap. The first strategy was to look for something projecting from the heap and to plan a way of grasping the projection which would not collide with other nearby objects. The next strategy was to take a grab at the whole heap. Finally Freddy would try to "bulldoze" through the heap. After each attempt, any object that had become recognisable would be "parked" in the reserved area.
After all the necessary parts had been found, a function was called to assemble the parts, using a force sensor to guide insertions.
Apart from a layer of control code, this project was implemented in POP-2.
This work used aspects of the entity-relation paradigm, with new contributions by Burstall and Barrow in the combinatorics of search.