To this day, manipulation still stands as one of the hardest challenges in robotics. In this work, we examine the board game Dr. Eureka as a benchmark to encourage further development in the field. The game consists of a race to solve a manipulation puzzle: reordering colored balls in transparent tubes, in which the solution requires planning, dexterity and agility. In this work, we present a robot (Tactical Hazardous Operations Robot 3) that can solve this problem, nicely integrating several classical and state-of-the-art techniques. We represent the puzzle states as graph and solve it as a shortest path problem, in addition to applying computer vision combined with precise motions to perform the manipulation. In this paper, we also present a customized implementation of YOLO (called YOLO-Dr. Eureka) and we implement an original neural network (NN)-based incremental solution to the inverse kinematics problem. We show that this NN outperforms the inverse of the Jacobian method for large step sizes. Albeit requiring more computation per control cycle, the larger steps allow for much larger movements per cycle. To evaluate the experiment, we perform trials against a human using the same set of initial conditions.

中文翻译：

---

尤里卡博士：人形机器人操纵案例研究

直到今天，操纵仍然是机器人技术中最艰巨的挑战之一。在这项工作中，我们将棋盘游戏 Dr. Eureka 作为鼓励该领域进一步发展的基准。游戏包括一场解决操纵难题的竞赛：在透明管中重新排列彩色球，其中的解决方案需要计划、灵巧和敏捷。在这项工作中，我们提出了一个可以解决这个问题的机器人（Tactical Hazardous Operations Robot 3），它很好地集成了几种经典和最先进的技术。我们将拼图状态表示为图形并将其解决为最短路径问题，此外还应用计算机视觉与精确运动相结合来执行操作。在本文中，我们还提出了一个定制的 YOLO 实现（称为 YOLO-Dr. Eureka），我们为逆运动学问题实现了一个基于原始神经网络（NN）的增量解决方案。我们表明，对于大步长，该 NN 优于 Jacobian 方法的逆方法。尽管每个控制周期需要更多的计算，但更大的步长允许每个周期更大的运动。为了评估实验，我们使用相同的初始条件对人类进行试验。