UR机械臂的“运动性奇异”

UR机械臂造型奇幻，做工精良，体型轻巧，令包括我在内的研究者一度惊叹其设计的精妙。
然而这两天我发现了其设计上的bug–“运动性奇异”

为了磨练写好的一些算法，我上周开始接触v-rep仿真软件（软件超级棒，连动量质量惯量这些物理参数都给模拟出来）。在该仿真软件里我选的是UR5机械臂作为模型。
经过一段时间的熟悉摸索，我用v-rep实现了既定路径的动态模拟，和实时控制的动态模拟。
动态模拟过程中，我突然发现，ur机械臂在某些特定位姿经常出现较大角度的扭动，经过分析，这不是算法有问题，而是机械臂的末端如果要保持某个朝向，当它经历某个路径时，它必然会有较大的身体调整来达到既定的位置点。 这个bug在球形腕机械臂上，是不存在的。
为了规避某些球形腕机械臂所具有的常规奇异点，UR的机械臂确实做到了更少的常规奇异位置，也就是说，它的确做到了很多球形腕机器人奇异的位姿都有解，问题在于当这些位姿连贯起来运动时，我们会发现它经常出现各角度值不连贯的情况。
因为我没有在任何文献中读到过对于UR机械臂的类似bug分析，我因此将这个bug命名为“运动性奇异”以区别其常规奇异点位。
根据笔者目前了解到的情况来看，UR机械臂的常规奇异点较少，一是分部基座z轴附近的柱状区域内（比如它无法做到在基座正上方的任意位姿）。二是五轴为0°时的多解奇异。

而“运动性奇异”是不同于上述两种常规性奇异的，常规性奇异比较好解决，有人用雅克比偏微分接近法，也有其它的一些方法来解决。而运动性奇异则真的无完美解，它是第四轴突出造成的结构性问题。

为了让大家更好的理解“运动性奇异”现象，再做如下对比说明：
在多解型奇异的情况下，假设奇异点在T点，如机器末端处于一个离该点较远的位置一次性（不用插值算法）通过六轴联动到达T点时，有无数种可能的各关节状态（一轴有两个可能状态可能，二三四六轴有无数的可能状态）；而“运动性奇异”点，如果要从远处通过插值的方式到达这个位姿，它可以和平常点位没有任何区别，就是八个解的位姿点位。但是和他相邻的点位却有巨大的六轴数值差距。

The UR robotic arm is fantastic, well-crafted and lightweight, making researchers including me once amazed at the exquisite design.
However, these two days I found a bug in its design – which I named “sports singularity”.

In order to sharpen some algorithms I wrote myself, I started to use the v-rep simulation software last week (the software is excellent, and the physical parameters like the momentum, mass ,etc. are vividly simulated). In this simulation software I chose the UR5 robotic arm as a model.
After a period of exploration, I used v-rep to achieve dynamic simulation of the established path as well as dynamic simulation of real-time control.
During the simulation process, I found that the UR robotic arms often have a wide range of twists while moving on its course. After analysis, I concluded that this is not an algorithm problem. If the end of the arm is about to maintain a certain orientation, when going through a path, the robot will inevitably have large-scale physical adjustments to reach a given point of the path. This bug does not exist on the spherical-wrist-Robots.
In order to avoid the conventional singular points of some spherical wrist manipulators, the UR’s mechanical arms do have fewer conventional singular positions by having a peculiar body construct. But the price is not very low, It is that when these poses move in a coherent motion,under some circumstances, the six angles can be inconsistent while it moves, what is more, the “sports singularity” points, vary from regular singular points, are not easy to locate, they show up when a certain orientation is maintained and the robot goes a long course.
Since I have not read a similar analysis for the UR robot in any of the literature, I thus suppose there has been no openly discussing for it. I would like to named the bug “sports singularity” to distinguish them from the regular singularity.
As far as I have learned, the conventional singular points of the UR robot arm are largely occurred in two scenarios , the first is the columnar region near the z-axis of the sub-base. The second is the multi-solution singularity when the fifth axe is on 0 or Pi.

The “sports singularity” is different from the above two conventional singularities. The conventional singularity is better solved. Some people use the Jacobian partial differential approach method, and there are other methods to solve it, too. The sporty singularity is it is a structural problem caused by the type of the fourth axis.
In order to let everyone better understand the “sports singularity” phenomenon, I would like tomake the following comparison:
In the case of multiple solution singularities, let’s assume that the singular point is at the T point. If the end of the machine is at a position far from the point (at no interpolation algorithm involved), the six-axis linkage reaches the T point, there are countless possible Joint state (there are two possible states for 1st axis, and there are countless possible states for the 2nd ,3rd,4th.and 6th axis). And for the “sports singularity” point, if you want to reach this posture from a distance(it can make no difference between the usual points, It can be the regular point with the eight solutions). We just can’t find a smaller(They are all too big to let the robot move smoothly) numerical difference solution(from the 8 solutions) from a certain posture adjacent to it and this point is just its next step on a path to go and unavoidable.