A 617Kb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE
I am treading down a dangerous path, speaking about design! My wife Nancy has a formal education in industrial design and often comments about my lack of ability in this department! Nevertheless, I venture forth. I hope by the end of this missive we circle back round to find relevance with LIDAR or related topics.
I think of optimal design as that point in the evolution of a design when it (the design, that is) is as close to perfect as it is going to get. Beyond this point, additions detract.
I recall when I saw the first version of the iPhone. The fellow showing it off to me was gushing with enthusiasm regarding the utility of the clean design elements. He was particularly enamored with the inertia-simulating roller setting on the alarm clock. I myself was dismayed by the lack of a stylus. I have always felt that Steve Jobs missed the point of optimization entirely when he stated (more or less) that humans come equipped with 10 perfectly good styli. If that were indeed optimal design we would be dipping our fingers in ink wells rather than participating in a multibillion dollar pen industry. I digress.
Probably one of the hardest things for an engineer is knowing when to stop! Some designs seem to have a natural stopping point since, to change the fundamentals, actually creates a new design. Of course, the most basic and useful fundamental design that comes to mind is the wheel. Optimize it past round and it is no longer a wheel (although attempts are made–Sun Tzu (sixth century BC) points out the utility of the four cornered stone for immobilizing equipment on hills).
There comes a point in design, I think, where one simply stops adding features (at least major ones) and focuses instead on usability of the features that already exist. This is certainly true for a lot of the software we produce in my own company. I visit our customers, asking about some new, really cool but complex feature that we have added to assist in production. I often get the reply "well, sounds nice but we have not been able to look at that one yet. But on this data import–it takes seven button clicks to do that. Can you reduce it to four?" Our customers need utility, of course, but they want simplicity to be front and center.
Take the ever-onward march toward higher and higher resolution with imaging systems (both cameras and LIDAR systems). Are we moving beyond the point of adding utility? This is an open-ended question and I certainly do not have the binary yes/no answer. Frame-based focal plane array (FPA) imaging is perhaps easy to answer. There is a tremendous industry drive for imaging chips, particularly of the CMOS realization. Thus as new chips come out in the prosumer and professional market, there is a holistic focus on design. As pixel density goes up, manufacturers are working very hard to improve the signal to noise ratio (SNR) or at least not let it degrade from generation to generation. This is a tough task since, for a given die size, increasing the pixel count decreases the individual pixel cell size.
While manufacturers of LIDAR systems are also in a density race, maintaining SNR in a LIDAR system is much more challenging than in an imaging FPA. First of all, the market for LIDAR sensors is orders of magnitude smaller than that for FPA imagers. This means there is much less money available to LIDAR designers for Research and Development. The physics are more difficult as well. Range precision (remember, precision is the measure of the spread the same measurement over repeated tries) is not getting significantly better (remaining at about 1 cm at 1 sigma). The second issue is position determination. This is pretty easy to back in to using block adjustment techniques with FPA technology, but we still do not have reliable recovery schemes for X, Y, Z, Pitch, Yaw, Roll with scanning sensors. In fact, this is the limiting factor in overall system precision. Note that after all these years of development, the GNSS/IMU systems are still totally segregated from the LIDAR unit itself with no real time interaction between the two.
Thus, we now are reaching the point where the density of LIDAR data is overlapping the precision ellipsoids. That is to say, our uncertainly in the position of a point is exceeding the spacing to the next point. The data are blurry when we zoom in close enough to discern individual points. Why continue down this path? This problem is severely exacerbated by the fact that no LIDAR hardware vendor yet offers a push-button geopositioning solution that provides a point cloud whose overall accuracy and precision can be reliably determined.
Thus perhaps it is time to step back and reassess where we are going with LIDAR technology. Have we reached the practical limits of scanning technology? Should our energy now be focused (pun intended) on getting FPA LIDAR to market? My own opinion is that we need to take a pause in one-upsmanship in the scanning LIDAR arena and really crack the calibration and geocoding nut. This would do more for the production industry than any other innovation.
Lewis Graham is the President and CTO of GeoCue Corporation. GeoCue is North America’s largest supplier of LIDAR production and workflow tools and consulting services for airborne and mobile laser scanning.
A 617Kb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE