Circular Saw Kickback Killer (We used science to make tools safer) - Smarter Every Day 209
Dustin and Chad (software expert) built a patent-pending multi-axis detection system that uses machine learning to identify circular saw kickback events and automatically apply the saw's built-in brake within milliseconds — before the user can physically react. The project, called Lantern Safety Kinetics, aims to get this technology adopted by major tool manufacturers. ---
Key Concepts
| Concept | Definition |
|---|---|
| Kickback | When a circular saw blade is pinched by bending wood, the blade grabs the material and violently propels the saw back toward the user |
| Multi-axis detection system | A 9-axis sensor array (accelerometer + gyroscope + magnetometer) embedded in the saw handle that monitors motion in real time |
| Machine learning (neural network) | Trained on data from normal saw use vs. kickback events; the algorithm determines what sensor patterns are unique to kickback — the programmer does not manually set thresholds |
| Dynamic motor braking | The saw's existing electromagnetic brake is triggered by the algorithm the moment a kickback signature is detected, decelerating the blade rapidly |
| Magnetometer trick | The magnetometer may detect the motor's own magnetic field, giving the algorithm additional signal about blade rotational state |
Notes
The Physics of Kickback
- Cutting sheet goods (e.g., plywood) causes cut sections to sag inward and pinch the blade
- Pinched blade behaves like a wheel: teeth catch the wood and launch the saw back at the user
- Blade fully exits the wood within **~80 milliseconds** of kickback initiation
- Once blade is freed, motor continues adding angular momentum → blade speed *increases*, making re-contact more dangerous
- Human reaction time is not fast enough to release the trigger before injury risk is realized
- The ideal intervention window is **before the saw leaves the wood**
High-Speed Camera Analysis
- Tests used a zip-tied trigger to observe full kickback behavior safely
- Timer overlay confirmed the entire dangerous event unfolds in under **one tenth of a second**
- Blade re-contacting a surface after kickback (e.g., falling back down) demonstrated severe secondary injury potential
Prototype v1 — Accelerometer Only
- Single-axis accelerometer detected backward acceleration of the saw
- Triggered motor shutoff automatically
- **Limitation**: false positives from normal bumps and handling
Prototype v2 — 9-Axis Sensor + Machine Learning
- Sensors: accelerometer + gyroscope + magnetometer (9 axes total)
- Training data: dozens to hundreds of normal use recordings + annotated kickback events
- Neural network identifies what sensor patterns are anomalous to kickback — no manual threshold programming
- Hardware embedded in a hogged-out saw handle
- Successfully stopped self-induced kickback in live hand-held tests
Reading the Sensor Data (Overlaid on Slow-Mo)
- **Yellow line**: overall acceleration magnitude of the saw body
- **Red line**: magnetic field strength (proxy for brake engagement and motor state)
- Sequence visible in data:
How the Brake Works
- The brake is **already built into consumer circular saws** (electromagnetic)
- The system only adds a sensor+algorithm layer to decide *when* to fire the brake
- No hardware modification to the braking mechanism itself is required
Actionable Takeaways
- If you want this technology in commercial tools, tweet the video at your preferred tool manufacturer (Dustin's explicit ask)
- Tool manufacturers can contact Lantern Safety Kinetics (link in original video description) to license or collaborate
- When using a circular saw, understand that kickback happens faster than human reaction time — use proper technique (blade depth, supported cuts) as the primary defense until active systems exist
Quotes Worth Keeping
The ideal time to respond to a kickback event is before the saw even leaves the wood.
I'm not really programming in exactly what the thresholds are — I'm taking data from dozens to hundreds of people using a saw in normal everyday ways and feeding that information into a neural network.
The brake's already in there. All you need is a little sensor to figure out when to hit the brake — and that's all we're doing.