The construction work and the machine action are very complicated. This is the main reason why the automation at construction site is very difficult and not realised yet. Focusing on the complicated and cooperative action of construction machinery, which needs combination of several actuators, some are repeated and some needs great skill. And all of them needs recognition of the environments as a result of work. Here, the necessity of cooperation between automatic control and human control are found. The fusion will be realised through recognising the intention which takes real shape in the manipulation and producing optimised controlled signals according to the prior knowledge. The coupled control system is proposed to realise the system in the blade control of the bulldozer, as a typical case of construction machinery. This system is consist of two elements. One is to produce optimised control signal for each job, and the other is to recognise intention of the operator from his manipulation and to select the optimised signal for the job. In order to realise the key unit of recognition, the neural network is selected. In application of the neural network, the problem of confusion in the trangent period between typical pattern was treated to application of the variogram. And the neural network with competitive learning system could recognise the levelling operation from the manipulation signal and the pitching signal.