The objective of our study is to develop a safety measure that ensures smooth control mode switching of an intelligent assist device (IAD)-Skill-Assist for the purpose of effective automotive manufacturing. Smooth switching, i.e., switching without abrupt deceleration, from the hands-off control mode (used for automatic travel) to the hands-on control mode (as typified by a power-assist control mode) is expected to enhance the productivity of IADs. However, the hazardous behaviors, i.e., careless behaviors that disregard an approaching IAD in hands-off control mode might cause collision and serious injuries to a human operator. For safety purposes, we focused on the operator's gesture of reaching for the force sensor of the Skill-Assist. Laser range sensors were employed to recognize the reaching gesture during control mode switching. A reaching gesture recognition algorithm, based on a hidden Markov model, processes characteristic patterns acquired by the laser range sensors, and judges the operator safety. Experimental results suggest that the Skill-Assist to which the proposed safety measure was applied could recognize the reaching gesture of an operator when the control mode was smoothly switched. In an experiment on hazardous behavior, the Skill-Assist could predict that the operator would be unsafe.