How We Built the Jigsaw Puzzle Solving Robot

Introduction

• Introducing Jigsaw, a robot built to solve jigsaw puzzles rapidly.
• Hope to be 200 times faster than the fastest human puzzler.
• Overview of the journey to the ultimate robot-human jigsaw puzzle face-off.
• Insights for human jigsaw puzzlers along the way.

Human Capabilities and Challenges in Puzzle Solving

• Humans perform four intricate tasks to solve puzzles:
  1. Picking up a piece: Human hands have 27 bones, 34 muscles; precise and dexterous.
  2. Rotating the piece: Simple for humans with our highly articulate arms.
  3. Moving the piece: Requires entire arm movement; humans have sophisticated arm capability.
  4. Deciding the piece’s position: Involves visual perception, spatial reasoning, visual memory, and executive function.
• Highlighting the special capabilities of human brains.

Building a Robot to Perform the Same Tasks

Step 1: Picking Up a Piece

• Replacing the thumb with a tiny specialized suction cup.
• Solenoid cuts off/connects to vacuum pump for precise suction control.

Step 2: Rotating the Piece

• Suction cup attached to a fine-tuned donut motor with accuracy down to 0.005 degrees.

Step 3: Moving the Piece

• Modified an avid CNC router with ClearPath industrial servo motors for high precision.
• Achieves accuracy down to .0005 inches.

Step 4: Deciding Where to Place the Piece

• The hardest problem due to complex human neural pathways.
• Simplified approach by ignoring puzzle print and focusing on edges.
• Used phone camera to take pictures, convert edges into splines for matching.
• Reduced solution space and evaluated fit by quantifying spline mismatch areas.

Key Challenges and Solutions

• Shane’s video on puzzle-solving machine challenges our progress.
• Collaboration with Ryan from Zipline led to breakthrough solution.
• Used edge analysis and serpentine pattern for photographic mapping.
• Solved matching with spline calculation and reduced unnecessary comparisons.
• Used a serpentine search strategy, backtracked on errors.

Scaling Up

• Successfully solved a 12-piece puzzle and scaled up to a 1000-piece puzzle.
• Dealt with compounded errors and small inaccuracies by approximating human tactile feedback.
• Added sensitivity to the z-height encoder for final adjustments.

Fit and Finish

• Ensured every piece snapped perfectly into place with the wiggle routine.
• Demonstrated actual solving, showing solving time and manual placement.
• Iterative modifications and testing led to successful 1000-piece puzzle solving.

Human vs. Robot Face-Off

Intermediate Matches

• Tammy McLeod vs. Kristen Bell in a 30-piece puzzle and a 500-piece puzzle with advantage.
• Tammy easily won by applying efficient strategies.

Human Tips

1. Dump out and turn all pieces over.
2. Reassess traditional edge-first approach based on context.
3. Group pieces by color, texture, or patterns.
4. Sort by shape for uniform-looking pieces like sky or plain areas.

Final Showdown

• Tammy faced off against the robot Jigsaw.
• Jigsaw showcased solving efficiency with periods of troubleshooting taken into account.
• Final victory for Jigsaw after human efforts supplemented by new technology.

Conclusion and Encouragement

• Demonstration of overcoming human limitations with advanced robotics.
• Encouragement for engaging with programmable robots via CrunchLabs HackPack for teenagers and adults.
• Promotion of the series to bridge the gap between entertainment and practical learning.