TAIR: a robot run by a neuron-like network
In September 1975, at IJCAI-75 in Tbilisi, Nikolai Amosov, Ernst Kussul and V. D. Fomenko of the Department of Biocybernetics at the Institute of Cybernetics in Kiev described TAIR, a transport robot controlled not by a program but by a neuron-like network of 100 nodes in six spheres, with a system of reinforcement and inhibition and 60 input channels from its sensors. The undercarriage has three wheels and measures 1600x1100x600 mm.
Why it matters
The Kiev approach, in which a robot's movement is governed not by a program but by a hardware network where each node stands for a notion and each connection for a relation between notions, reached an international AI conference with its architecture, sensors and test plan. The authors themselves say they are reviving the network approach after unsatisfactory attempts and that its competitiveness is not yet shown.
The department began designing the robot in 1972. A network of 26 nodes and about 300 connections was checked only on a simulator, in a hypothetical environment on a map; the moving part, the lower-level control and individual sensors were tested in autumn 1973, and experiments with the whole system were scheduled to the end of 1975. The robot is given a map of the area in advance. Next came plans for a network of 1000 nodes and 2x10^6 learning connections, speech contact, manipulation and vision. What the record does not claim: that the robot moved autonomously in a real environment by the time of the paper. The research report behind the candidate said so; the paper describes those tests as planned. The report also gave pages 722-726; the page footers read 727-730.