coneblog

So, There I sit, hands poised above the Skycam computer keyboard, quaking like an Aspen. Garrett nods, and I press enter to start the control loop. Nothing happens (which is good!) I nod back to Garrett, and he flips the Brake switch to “Off”. There is a loud whack sound of the brakes releasing, and…

…And nothing. Which is very good. The system is live, and stable at its origin position, with no control inputs. Garrett eases forward ever-so-lightly on the Z-stick, which controls vertical movement. The weight slowly rises a few feet. He eases back, and it lowers a few feet. Up and down a few times, and a weak cheer rises. The system is moving, the lines are tight, the motion calculations are working.

But, in a few minutes, it is clear that all is not well. As the weight moves horizontally away from the origin in the center, the lines begin to slack. The farther away from the center, the more slack in one or two cables. All eyes turned to me – this was clearly a software problem, as in all other respects the system worked fine.

I moved the unit around the gym, and tried to understand the error pattern. In the center, the lines were taught and perfect at any height. The error pattern was bilaterally symmetric; meaning that the pattern of slack lines was identical in each quadrant of the rectangular basketball court. Along the center axis, the two longest lines were slack. Moving diagonally towards any corner pulley, that corner line was tight, the opposite corner was very slack, and the two side cables were equally, slightly slack. Plus, the payload tended to dip toward the floor as it moved towards the edges of the pattern.

Clearly, something non-random was occurring here, but I couldn’t see what. The system is nothing but triangles- what was going wrong? Why did it work in the center? Had my hand-coded, Newton’s method square root routine sprung a leak? Had the Pythagorean Theorem finally given out? Any Ideas?