We describe in this paper, the development of a humanoid robot for the 2018 Ski Robot Challenge, which was held in Korea. The goal of this competition was to develop a humanoid robot that could autonomously perform in a ski giant slalom game. To participate in this competition, we developed the humanoid robot RoK-2 which is designed to ski on a slope. The lower body joints of the robot were developed to have high torque because it needed to perform ski motions in a situation where shock and vibration would occur. The upper body of the robot was designed to be shielded from factors that could damage electronic components such as snow. The ski robot required a ski turning mechanism so that it could make its way between the ski flags on the slope. To perform the turns accurately, a ski plate for the desired radius was selected and the robot performed a proper, humanlike motion using the extension–flexion a motion of carved turn. To verify the stability of the motion, we confirmed that the zero-moment point (ZMP) of the robot was within a support polygon of the robot. The environment of a ski slope is not good for image processing because of the changing light and weather. To overcome this problem, a deep learning based ski flag recognition algorithm was developed so that the robot could recognize ski flags correctly. Finally, RoK-2 was verified in the 2018 Ski Robot Challenge.

我们在本文中描述了为在韩国举行的2018年滑雪机器人挑战赛开发的人形机器人。这项比赛的目的是开发一种可以在滑雪大回转比赛中自主表演的人形机器人。为了参加比赛，我们开发了人形机器人RoK-2，该机器人旨在在斜坡上滑雪。机器人的下半关节被开发为具有高扭矩，因为它需要在可能发生冲击和振动的情况下执行滑雪运动。机器人的上身被设计成不受可能损坏电子组件（例如雪）的因素的影响。滑雪机​​器人需要一个滑雪转向机构，以便可以在斜坡上的滑雪标志之间移动。为了准确进行转弯，选择了所需半径的滑雪板，然后机器人使用伸展弯曲弯曲的动作进行了适当的，类似人的运动。为了验证运动的稳定性，我们确认了机器人的零力矩点（ZMP）在机器人的支撑多边形内。由于光线和天气的变化，滑雪场的环境不利于图像处理。为了克服这个问题，开发了一种基于深度学习的滑雪标志识别算法，以便机器人可以正确识别滑雪标志。最终，RoK-2在2018年滑雪机器人挑战赛中得到验证。由于光线和天气的变化，滑雪场的环境不利于图像处理。为了克服这个问题，开发了一种基于深度学习的滑雪标志识别算法，以便机器人可以正确识别滑雪标志。最终，RoK-2在2018年滑雪机器人挑战赛中得到验证。由于光线和天气的变化，滑雪场的环境不利于图像处理。为了克服这个问题，开发了一种基于深度学习的滑雪标志识别算法，以便机器人可以正确识别滑雪标志。最终，RoK-2在2018年滑雪机器人挑战赛中得到验证。