Force control design for robots based on correlation in human demonstration data

Abstract

This paper proposes a method to construct a force control law for industrial robots by using the normalized cross-correlation (NCC) of human demonstration data. Conventionally there are two solutions where one is based on human demonstration data, and the other is based on numerical optimization. The former one gives the force control parameters efficiently, but the parameters may make the robot unstable. On the other hand, the latter can maximize the performance, but it requires an enormous number of trial and error. Taking into account the merit of each method, we consider combining the two approaches. In our proposed method, the force control laws are determined to maximize the NCC of the human demonstration data. The orientation of the coordinate system is also determined to maximize the NCC of the human demonstration data. Then, the parameters included in the force control law are optimized by the downhill simplex method. The proposed method was applied to a ring-shaped rubber packing assembly task, and it could realize an assembly with human-like performance. Moreover, some comparable experiments were also conducted, and it was confirmed that the proposed method can construct appropriate force controls.

Keywords

• industrial robot
• force control
• programming by demonstration
• cross-correlation
• optimization
• assembly
• flexible object
• ring part