日前，媒体报道了这样一个案例：一家制造企业的人力资源负责人说，今年2月，公司去一家985大学招聘机械、液压专业学生，给出的月薪高达1.4万元，“我们下了很大的决心，才给出这样的标准，是公司历史上从未有过的”。但是，这次专场招聘竟无人问津，“一个学生都没走进来”。后来一打听，这一届的学生早已被其他大厂以更高的工资提前签约锁定，“人家一开口就是年薪二十几万三十几万”。

一方面，是社会整体的就业压力，另一方面，是制造业普遍缺少技术工人，这种结构性就业矛盾，其实并非新鲜事。近年来，随着制造业升级，对于技术工人，尤其是高级技工的需求进一步增大，这种矛盾更是有加剧的趋势。

前言

水下机器人技术不断发展，应用不断增加，仿生机器鱼作为水下机器人中的一种，获得越来越多的关注。仿生机器鱼同时作为一种仿生学的产物，具有很多先天的优势。例如，运行噪音小，体积小，灵活性高，可以很好的融入水下生态环境。

一、仿生软体机器鱼的机械设计与材料选择

优良的机械设计是仿生机器人的应用基础。

These days I have studied unmanned driving path planning and reinforcement learning, and have gained some preliminary knowledges.

In reinforcement learning, there are some simple models, such as letting the agents find a way out of the maze, which may be similar to unmanned driving path planning. To do this kind of reinforcement learning problems, the most used way is try-and-error. That is a lot of attempts, according to the final return (find the exit or not) to update themselves in a certain state (position). Then the agents select a certain action (moving forward, backward, left, and right), so as to explore an optimal policy.

The advantage of using reinforcement learning to do this is that you only need to tell the agents your purpose (find the way out), without telling them how to do it. As long as you try more times, the agent can always find a more ideal strategy (the value function converges), which is the path of action. The specific algorithms can be Monte Carlo or Q-learning. You also can try the Dyna-Q algorithm.

This is another SOM based formation control code in 2-D and 3-D space. Please refer to Github

Email: lixin850224@163.com

This is my preliminary codes using SOM method (Self-organizing Map Neural Network) to help you with your own formation control research. Please refer to Github

Authors: Xin Li

Email: lixin850224@163.com

Shanghai Maritime University

百年征程波瀾壯闊，百年初心曆久彌堅。“100”的背後是什麼？是不忘初心，是唯有奮鬥，是一個你我接過時代的接力棒，奮勇直前。一百年恰是風華正茂，新征程仍需砥礪前行。