이 논문에서는 로봇 팔의 말단 장치 경로 계획을 위해 Catmull-Rom 곡선을 기반으로 한 MoveCR 함수를 개발하였다. 이 함수는 CRI를 바탕으로 학습된 궤적의 문제점을 보완하기 위해 설계되었다. 강화 학습으로 생성된 궤적은 끝부분으로 갈수록 떨림 현상이 증가하여 로봇의 성능을 저하시킬 가능성이 있다. 이를 해결하기 위해 개발된 MoveCR을 활용하여, 떨림 현상이 발생하는 궤적 구간을 MoveCR 기반의 궤적으로 대체하는 데 성공했다. MoveCR 궤적은 떨림 현상이 없으며, 불연속점이 없는 매끄러운 경로를 제공한다. 따라서 강화 학습과 MoveCR을 결합하면 로봇 팔의 부드럽고 최적화된 안전 경로를 설계할 수 있다.

With the growing demand for robots in the industrial field, robot-related technologies with various functions have been introduced. One notable development is the implementation of robots that operate in collaboration with human workers to share tasks, without the need of any physical barriers such as safety fences. The realization of such collaborative operations in practice necessitates the assurance of safety if humans and robots collide. Thus, it is important to establish criteria for such collision scenarios to ensure robot safety and prevent injuries. Collision safety must be ensured in both pinching (quasi-static contact) and impact (transient contact) situations. To this end, we measured the force pain thresholds associated with impacts and evaluated the biomechanical limitations. This measurements were obtained through clinical trials involving physical collisions between human subjects and a device designed for generating impacts, and the force pain thresholds associated with transient collisions between humans and robots were analyzed. Specifically, the force pain threshold was measured at two different locations on the bodies of 37 adults aged 19-32 years, using two impactors with different shapes. The force pain threshold was compared with the results of other relevant studies. The results can help identify biomechanical limitations in a precise and reliable manner to ensure the safety of robots in collaborative applications.