Week 6#
Roo bars, AEB, and their blind spots#
The Brochure Has Excellent Daytime Vision#
A polished claim#
“Our system detects wildlife before you do.”
Which wildlife? At what distance? In what lighting? With what consequence?
A claim that answers none of these hasn’t told you anything yet.
Four different functions#
- detection — identifies a possible hazard
- warning — alerts the driver
- automatic emergency braking (AEB) — changes the vehicle’s motion
- impact protection — changes the collision, if one occurs
A roo bar is impact protection only. It does not detect. It does not brake.
False positives, false negatives#
100 staged trials: 40 with an animal present, 60 without.
| Trial | Detected | Missed |
|---|---|---|
| Animal present (40) | 34 | 6 |
| Trial | False alarm | Correctly quiet |
|---|---|---|
| No animal (60) | 9 | 51 |
False negative rate = 6/40 = 15%. False positive rate = 9/60 = 15%.
Different denominators. A single “accuracy” figure hides both.
Detection distance meets the margin#
Distance margin = assumed detection distance − modelled stopping distance
- 120 m stated range, 80 km/h (71.3 m stopping): margin +48.7 m
- same system, 60 m once obscured by vegetation: margin −11.3 m
The headline range and the fine-print range are not the same claim.
Detection rate ≠ collision prevention#
85% detection is not 85% of collisions prevented.
Between detection and prevention: warning comprehension, driver reaction, whether the vehicle can actually stop in time, whether the animal stays put.
One is a claim about a sensor. The other is a claim about an entire chain.
Performance varies by condition#
- lighting — daylight, dusk, night, glare
- animal type and posture — stationary and broadside vs. fast and end-on
- obstruction — vegetation hides animals from sensors too
- speed — validated at 60 km/h isn’t validated at 100 km/h
A card that states one condition has told you about one condition.
Three cards and a roo bar#
| Card | Function | Range/distance | Latency | Unverified for |
|---|---|---|---|---|
| A — optical | detect + warn | 90 m daylight, degrades below ~20 lux | 0.4 s | low light |
| B — thermal | detect + warn | 120 m, ~60 m if obscured | 0.6 s | reptiles, small animals |
| C — AEB fusion | detect + brake | TTC-triggered, to 60 km/h | 0.3 s | wildlife, >60 km/h |
| Roo bar | impact protection | n/a | n/a | compatibility, injury outcomes |
Designing a fair test#
A test can favour a product without lying — just by choosing conditions.
A fair test states conditions in advance, includes the ones a vendor would rather skip, reports a real denominator for every rate, and says plainly what it didn’t test.
Insufficient evidence is a valid answer#
None of the three fictional systems has been shown to prevent a collision.
Each has been shown to do something narrower, under stated conditions.
“Conditionally promising” is an honest ceiling. You are not required to pick a winner.
Where this connects to your assessment#
Checking a denominator. Separating detection from prevention. Reading a specification’s stated limits.
The literature review and hotspot analysis — due 12 April 2027, 40% — rewards exactly these habits, applied to published research.
“Wildlife detection” is a category. “The kangaroo behind that bush” is a test case.
The brochure has excellent daytime vision.