Tutorial · 10 minutes
Read a recommendation
Every worked example is one page: the answer, the runs, the reason for each choice, and what to build. This guide reads the returns tote page part by part, with its own data.
01The answer
Returns tote
1A single industrial servo parallel gripper, 110 mm stroke, finishes the task in 8 of 8 runs; it is also the best single off-the-shelf tool.
2Rigid3Simulated, verify on hardwareProduct mode: not part of the pre-registered experiment
- 1One sentence firstWhat to use, how often it finished, and how the best single off-the-shelf tool did on the same runs.
- 2FidelityRigid: A rigid multibody model with a scripted controller.
- 3Two labels on every answerThese are simulation rankings under stated conditions, not hardware results, and not the pre-registered experiment.
02The replays
Both runs are recorded physics on the same condition, played on one clock. Press play, drag the scrubber, or click a step on the timeline to jump both replays there.
t 00.00 / 00.00 s
Loading replay
Recommended: Industrial servo parallel gripper, 110 mm stroke, nominal placement, seed 0. Completed in 78.2 s.
Recommendeddone at 78.2 s
Click a step to jump both replays to it. Hatched gaps are a head switch or tool change: a fixed time, not simulated.
- 1Two lanesThe recommended setup on the left, the best single tool on the right, same placement and seed.
- 2Changing toolA tool change or head switch is a fixed time, not simulated; the replay holds still under a "Changing tool" card instead of animating it.
- 3The step timelineEach block is a step, green when it passed and red when it failed; hatched gaps are tool changes.
- 4Where it stopsA run that fails ends at a red line, with the reason.
03A step card
One card per step says what the step takes, how many tools could do it, which one the recommendation uses and why. This is step 3, the one the fewest tools could do. Its “Watch this step” button jumps the replays above to it.
Step 3 of 4
Pick the pack
Pick the pack and place it in the tote.
- Skill
- Grasp
- Object
- Shrink-wrapped pack
- Into
- Tote
- Done when
- ends inside the target; tilt at most 30 deg
4 of 18tools tried in physics could pick the pack
38 more of the 56 were ruled out before physics; 54 screen runs in all.
- 4passed
- 1passed some runs
- 13fell short of the goal
- 38ruled out before physics
Best single tool: Passes in 8 of 8 runs.
The tool, kept from step 1
Industrial servo parallel gripper, 110 mm stroke
100% of full-task runs passed this stepRigid
Already on the wrist, so there is no change here; it was chosen at step 1 for every step it covers.
See which tools (56)
Passed every screen run (4)
- Industrial servo parallel gripper, 110 mm strokeIndustrial gripper, the recommendation's pick
- Suction cup, 16 mm bellowsSuction cup, 3 of 3 runs
- Suction cup, 25 mm bellowsSuction cup, 3 of 3 runs
- Suction cup, 40 mm bellowsSuction cup, 3 of 3 runs
Tried and failed (14)
- Suction cup, 10 mm bellowsSuction cup, 2 of 3 runs, fell short of the goal
- Angular-jaw gripper of an open hobby armIndustrial gripper, 0 of 3 runs, fell short of the goal
- Rack-and-pinion parallel gripper for an open hobby armIndustrial gripper, 0 of 3 runs, fell short of the goal
- Two-finger parallel gripper, teleop-arm fingersIndustrial gripper, 0 of 3 runs, fell short of the goal
- Five-finger 18-actuator hand, pinchMulti-finger hand, 0 of 3 runs, fell short of the goal
- Five-finger 18-actuator hand, power graspMulti-finger hand, 0 of 3 runs, fell short of the goal
- Five-finger 22-joint hand, pinchMulti-finger hand, 0 of 3 runs, fell short of the goal
- Five-finger 22-joint hand, power graspMulti-finger hand, 0 of 3 runs, fell short of the goal
- Four-finger 16-joint hand, pinchMulti-finger hand, 0 of 3 runs, fell short of the goal
- Four-finger 16-joint hand, power graspMulti-finger hand, 0 of 3 runs, fell short of the goal
- Servo-driven four-finger hand, 13 actuators, pinchMulti-finger hand, 0 of 3 runs, fell short of the goal
- Servo-driven four-finger hand, 13 actuators, power graspMulti-finger hand, 0 of 3 runs, fell short of the goal
- Tendon-driven five-finger hand, 16 joints, pinchMulti-finger hand, 0 of 3 runs, fell short of the goal
- Tendon-driven five-finger hand, 16 joints, power graspMulti-finger hand, 0 of 3 runs, fell short of the goal
Ruled out before physics (38)
opens narrower than the part (11)
- Two-finger adaptive linkage gripper, 85 mmIndustrial gripper
- Two-finger linkage gripper of a cobot armIndustrial gripper
- Force-controlled parallel gripper, 48 mm strokeIndustrial gripper
- Two-finger parallel cobot hand, 80 mmIndustrial gripper
- Two-finger parallel gripper, rigid pads AIndustrial gripper
- Two-finger adaptive gripper, pinchIndustrial gripper
- Two-finger adaptive gripper, power graspIndustrial gripper
- Air-actuated mini parallel gripperIndustrial gripper
- Three-finger adaptive gripper, pinchIndustrial gripper
- Three-finger adaptive gripper, power graspIndustrial gripper
- Two-finger parallel gripper, fin-ray padsIndustrial gripper
digits stand too close for the part (9)
- Designed four-finger swinging hook gripper 052cDesigned morphology
- Designed three-finger sliding hook gripper da49Designed morphology
- Designed three-finger hand, 2 joints per finger d55eDesigned morphology
- Designed four-finger hand, 3 joints per finger b56aDesigned morphology
- Designed three-digit segmented tentacle f293Designed morphology
- Designed three-finger hand, 2 joints per finger 3e58Designed morphology
- Designed four-finger hand, 1 joint per finger e352Designed morphology
- Designed three-finger hand, 1 joint per finger ca8bDesigned morphology
- Designed three-digit segmented tentacle 90e8Designed morphology
can't be built for this task (5)
- Parallel jaw, configurator imprint: Small cartonParallel jaw
- Parallel jaw, flat pad: Small cartonParallel jaw
- Parallel jaw, configurator imprint: Shrink-wrapped packParallel jaw
- Parallel jaw, flat pad: Shrink-wrapped packParallel jaw
- Parallel jaw, stock blade: Shrink-wrapped packParallel jaw
can't pick or hold a part (4)
- Button-pusher toolIndustrial gripper
- Drawer-pull hook toolIndustrial gripper
- Spring-loaded marker holder toolIndustrial gripper
- Tamper pad, 200 mmProcess toolrated 40 N (the stage slips at 60 N)
the part is wider than the arm reaches (3)
- Soft arm, 12 segments, 140 mmSoft arm
- Soft arm, 16 segments, 160 mmSoft arm
- Soft arm, 20 soft segments, 180 mmSoft arm
the 150 x 100 mm top face of pack covers 0 of the frame's... (2)
- Vacuum area frame, 2 x 2 cups, 300 mmVacuum area framerated 4.15 kg on a smooth face (holding force over a 1.5 safety factor)
- Vacuum area frame, 2 x 3 cups, 600 mmVacuum area framerated 8.47 kg on a smooth face (holding force over a 1.5 safety factor)
a nut runner turns a fastener; it is not a hook, so it ca... (1)
- Nut runner, 13 mm hex socketProcess toolrated 6.7 N m (clutch cap 10 N m); members at least 12 mm tall
heavier than the arm can carry (1)
- Three-finger 12-joint industrial handMulti-finger hand
jaws shaped for a different object (1)
- Parallel jaw, stock blade: Small cartonParallel jaw
measured: its tip reaches 0 mm sideways past the rest of ... (1)
- Servo screwdriver attachment for a hobby armIndustrial gripper
- 1What it takesThe skill, the object, the limit and the test that says the step is done.
- 2Who could do it4 of 18 tools tried in physics passed every screen run; the bar splits the whole field of 56 by outcome, and tools that could not do the step at all were ruled out first, with a reason.
- 3The pick, and whyCounted from the success matrix, the candidates and the finalists. Where the data cannot separate two tools, the card says so.
- 4The baseline on this stepHow the best single tool did here, and on the held-out conditions when the benchmark covered it.
04Head to head
Recommended
Industrial servo parallel gripper, 110 mm stroke
Best single off-the-shelf
Industrial servo parallel gripper, 110 mm stroke
- Finishes the task
- 8 of 8
- 8 of 8
- Cycle time
- 78.2 s
- 78.2 s
- Tools
- 1 single tool
- 1 single tool
- Tool mass
- 1.09 kg
- 1.09 kg
- Fidelity
- Rigid
- Rigid
- 1Same conditionsBoth setups ran the same 8 conditions and seeds.
- 2Cost of the answerCycle time, number of tools and their mass: what the extra completion costs.
- 3Held-out checkOn the story page, a pre-declared benchmark adds new conditions neither setup was chosen on, when one covered the use case.
05The build pack
One download with everything to build and check the setup. Every file states its licence.
For the recommended setup
Industrial servo parallel gripper, 110 mm stroke
7 files in one zip, every one under an open licence. Print the parts, buy the tools on the bill of materials, and check the result on your hardware before you rely on it.
Download the build packzip, 10.8 kB, sha256 635bcdee89e4
Bill of materials
What to buy or print, with quantities.
bom.csv
Assembly notes
How it goes together, and a read-me.
ASSEMBLY.md, README.txt
Simulation scenes
The tested scenes, to re-run or extend.
mjcf/setup_tool0.xml
Certificate
The runs behind the recommendation.
certificate.html
Credits
Licences of everything inside.
CREDITS.txt, licences/tool0_LICENSE.txt
06When nothing finishes
Some jobs have no setup that completes them. Then the page says so first, shows the closest setup in place of a recommendation, with its replay and step cards, and names the step every finalist stops at and how far each tool family got against the target.
Read the caveats