Collision detection and reaction control is the most important function of a manipulator which cooperate with a human to prevent the injury caused by an unexpected collision. Intuitively, the difference between the input torque of manipulator and actual joint torque which is calculated from the inverse dynamic model of manipulator could be able to represent disturbances. However, the point is collision distinction among the various disturbance signal. Especially, the disturbance signal caused by modeling error which is the main causes of fault detection. In this reason, we are studying about the improvement of this inverse model-based collision detection algorithm.

The computed disturbance torque signal is filtered with a frequency filter which has special bandwidth related with manipulator structure. In this manner, the error signal caused by modeling error is successfully reduced. Even the hydraulic manipulator with massive actuation torque and modeling error, the collision is successfully detected. The bandwidth can be calculated with simple theory at any type of S-R-S formed manipulator. The developed methods are verified with collision simulation and experiment.

Also, various other research topics related to manipualtor control is going on : the manipulator design and analysis, joint torque sensing method, collision reaction control.