Each button opens one page. Blue = the real robot. Green = a simulated run, with the simulator's ground truth.
| 1 | Sat Oct 24 | day one: slides in, lifts a cube 80 mm | real video |
| 2 | Wed Oct 28 18:16 | moves the red cube to the left | real video + what it saw |
| 3 | Wed Oct 28 20:26 | blue cube into the blue bin, then checks its work | real video + what it saw |
| 4 | Wed Oct 28 20:34 | cubes by color: one in, two grabs missed | real video + what it saw |
| 5 | Wed Oct 28 evening | cubes by color: two in, then a fault | real video + what it saw |
| 6 | Oct 29-30 | two cubes placed, faults on the third | real video |
| 7 | Oct 29-30 | blue cube into the blue bin, first grab | real video |
| 8 | Fri Oct 30, ~5:40 pm | first pen moved (a slide grab) | real video |
| 9 | Fri Oct 30, ~8 pm | black pen moved | real video |
| 10 | Fri Oct 30, ~8 pm | red pen into the red bin, first grab | real video + what it saw |
| 11 | Fri Oct 30, ~8 pm | blue cube into the blue bin, first grab | real video + what it saw |
| 12 | Fri Oct 30 20:09 | two pens moved, then one set down on another | real video + what it saw |
| 13 | Fri Oct 30, ~8 pm | two pens moved, the second crushed | real video |
| 14 | Fri Oct 30, ~8 pm | pens grouped: three clean grabs | real video |
| Every recorded try on the real cell (60), with all three cameras | browse all footage |
| a | In simulation, depth is clean | picture |
| b | On the real cell, the depth sensor bubbles | 3D recording |
| c | What it saw seconds before a crush | picture |
| d | The same cube, measured 15 times | chart |
| e | Tried: fog of war (many frames voting) | 3D recording |
| f | Tried: structure from motion (tracking points) | 3D recording |
| g | ... sub-millimetre in sim, scattered on real | picture |
| h | What worked: measure against the table right next to it | picture |
| Every recorded try on the real cell (60), with all three cameras | browse all footage |