In recent years, most countries have experienced a shift toward aging population, leading to a decline in the working population; thus, an increased physical burden on workers results. The use of powered exoskeletons is one approach to reduce physical stresses on the bodies of workers and elderly people. However, conventional powered exoskeletons have many sensors and actuators mounted on them, resulting in problems associated with their high price, high weight, etc. In this paper, we introduce a new high-efficiency reduction gearbox, which is being researched and developed in our laboratory as a substitute for conventional actuators, and present the results of its attempted assist control without a torque sensor. It was found that the user's torque can be estimated with high accuracy without using a torque sensor by using a highly efficient gearbox with 1/102.14 reduction ratio. In addition, we performed motion identification using a nonlinear support vector machine with the aim of assist control with reduced discomfort on the body by suppressing vibrations.