# [2021-11-26] Mr. Thijs Roumen, Hasso Plattner Institute, "portable laser cutting"(English Speech)

As computation power grows up, many machine in laboratory reaches consumers such as computer, 3D printer, desktop, digital video and audio. How about laser cutting? Our main question is that what does it take for laser cutting to reach the specific potential such as to reach consumers? His working hypothesis is portability.

The portability of laser cutting means the ability to reproduce models independent of the laser cutter and to allow users to advance models by building on the work of others. Let’s look at portability in other field such as computing. Initally, the program is written for expensive hardware and when the hardware changes, the program cannot run. Nowadays, the program is written for multiple platforms. In the coding view, there is also the trend from assembly code to cross-platform higher order programming language when the compiler shows up. Hence, the portability of computing is the ability of software to be transferred from one machine or system to another.

So the standard is that we have to establish norm or requirement for a repeatable technical task. The definition of portability changes as well. In means that the ability of something to be transferred from one machine or system to another.

So what is preventing portability in laser cutting today? Is laser cutting really pre-portable? If so, how can we make it portable? The 2D cutting plan are already pretty portable. The idea of encoding laser cut models as 2D drawings has already come some way. But is has its limitation as well. The third part of laser cutter is necessary to edit like WORD/TeX for printing. Hence, the first attempt is to patch the sharing infrastructure of 2D cutting plans to reach low level portability. The author wrote a semi-automatic rewriter that converts a machine-specific cutting plan to a machine-independent plan. With KerfCanceler, he can make joints and moving part portable but fail to make non-portable joint changeable in kerf. Hence, he redesigns these non-portable part portable with spring fit. Spring Fit is a software tool that detects joints and mounts in models and replaces them with portable equivalents that are tolerant too variation in kerf. What about bearings, which do not tightly hold something? It can be solved in thec utting plan as well with kerf-canceling mechanisms, which is mechanisms that continue to perform their functionality independent of the amount of material removed by the fabrication machine.
So, how did we do by now? The resulting models did cut on any cutter, so we had achieved portability. But more importantly we wanted to edit the third part of the laser cutter. For example, if I wanted to adjust the focal range of the VR headset, we have to find out which 2D plates have to be modified. Then, we have to had that new model truly in 3D, the flatFitFab and kyub can help us visualiizing a virtual 3D object. Hence, three different parts are all worked great now and everyone could edit and build on it.
But there are still a lot of great models encoded as 2D cutting plans out there, which means 2D cutting plans are still the issue. The standard is to establish norm or requirement for a repeatable technical task but standards tend to stick somewhere. We need assembler to combine 2D plans to 3D. The assemble is a 5 step algorithm that converts 2D laser cutting formats into 3D format. The five steps contain plate detection, joint detection, material thickness detection, joint matching and semi-automatic assembly. With the match of joint, we can reduce the time complexity form O(n2) to O(n). For the first four steps, it is nearly 98% accuracy to combine the joint within average 107 ms. How about semi-assembly? We need an autoAssembler which is a heuristic function of 6 assembly heuristics, and an implementation that beam searches the exponential space of combinations of plates using that function. The hash table limits the fanout to joints that fit to the current candidate, resulting in much better performance as well. There are six heuristics to assess quality of assembly as well including preference for compact candidates, avoid intersections between plates, preference for unambiguous joints, minimize number of unmatched joints, use constraints posed by other plates, favor symmetric candidate. In the end of the heuristic, the algorithm optimize the hyper-parameter with the tree pazen estimator.

## Note
### The note I write is totally summarized version of speaker with minor my opinion. The citation is described below.

## Citation
### Topic: portable laser cutting
### Speaker: Mr. Thijs Roumen