This paper describes a walking control scheme for a quadruped robot. In conventional walking control scheme, the leg motion is generated from the robot moving trajectory beforehand, and it is difficult to control in real time. The proposed control scheme employs a real time command from a human operator. The operator gives a current planar moving velocity command to the robot. Since the command in the future cannot be predicted, the scheme does not generate a periodic gait. Body propulsive action is basically continued according to the command unless a problem arises. When a foot reaches the border of the work space of the leg, the body propulsive action is interrupted, and a supporting foot pattern is changed according to the leg transition rule. That is, a foot arrival at the border is considered as an event to trigger an recovery action for the body propulsion. The actions are decided to converge to the intermittent crawl gait when straight walking command is continued. The proposed control scheme is evaluated using actual walking robot.