Understanding Collision Evasion

Documented distinction

A warning informs the driver. Braking assistance changes vehicle speed. Steering assistance changes the path. NHTSA distinguishes these functions even when several appear in the same car. A dashboard message, a brake-pedal movement and a steering input are therefore different evidence. Recording which occurred is more useful than calling every response “self-driving”. [2]

Vehicle-specific fact

Tesla’s Collision Avoidance Assist chapter separately describes Forward Collision Warning, Automatic Emergency Braking and Obstacle-Aware Acceleration. The last can reduce acceleration near a detected obstacle; it is not interchangeable with emergency braking. The same chapter’s approximate 5–200 km/h FCW/AEB operating range belongs to the specified US Model Y manual. It is not a guarantee for every model, direction of travel, software release or road condition. [1]

Worked example

Suppose two vehicles are travelling in the same direction. In a simplified constant-speed model, closing speed is the following vehicle’s speed minus the lead vehicle’s speed. If a follower is at 60 km/h and the vehicle ahead is at 40 km/h, the gap is shrinking at 20 km/h, or about 5.56 m/s. With a 20-metre gap, the constant-speed time to contact is about 3.6 seconds. These are hypothetical inputs, not a detection or activation threshold.

Explanation

Change the lead vehicle to a stationary object and the same follower closes at 16.67 m/s. At a 20-metre gap, the constant-speed time becomes about 1.2 seconds. Neither number predicts what a particular assistance system will do. Braking by either vehicle, a cut-in or a curved path changes the calculation. This is why a single speedometer reading cannot describe the complete collision problem.

Engineering model

Time to contact is a model, not a certainty. It depends on the assumption that relative motion continues as estimated. A real system must also decide whether an object is in the relevant path, whether its motion estimate is reliable and what intervention is appropriate. That conceptual chain explains why detecting something in an image and deciding to brake are separate tasks.

Physics context

Stopping distance includes travel before braking begins and travel while speed is being reduced. Rhode Island’s driver guidance separates perception, reaction and braking, and notes that traction changes braking distance. Assistance may change parts of the response, but it does not remove the physical distance required to slow the vehicle. [6]

Worked example

For an educational constant-deceleration model, distance = v × t + v²/(2a). Here v is initial speed in m/s, t an assumed delay in seconds and a a positive constant braking deceleration in m/s². The first term grows directly with speed; the second grows with its square. Doubling speed therefore doubles the delay distance and quadruples the braking distance when the assumptions stay the same.

Explanation

The chart assumes a one-second delay and 6 m/s² constant deceleration on a straight, level path. It omits tyre variation, grade, brake build-up, changing friction and a moving target. At 60 km/h, delay travel is 16.7 metres and modelled braking travel is 23.1 metres, totalling 39.8 metres. These are arithmetic results from chosen inputs, not measured Tesla stopping distances, safe following gaps or AEB performance specifications.

Primary research

Jessica Cicchino’s 2017 IIHS study compared police-reported crashes in 22 US states during 2010–2014. It estimated rear-end striking-crash reductions of 27% for forward collision warning alone, 43% for low-speed AEB and 50% for warning with AEB. The corresponding injury-crash reductions were 20%, 45% and 56%. These describe the studied systems and populations, not Tesla’s current FSD or a Collision Evasion release. [4]

Explanation

Those percentages are relative reductions in crash-involvement rates, not a guarantee that an individual intervention succeeds. The population, exposure measure, comparison group and crash definition matter. A result for rear-end crashes should not be relabelled as a result for all crashes, pedestrian incidents or autonomous driving.

Primary research

A 2019 IIHS study examined the rear-end crashes that remained. Turning, lane-changing lead vehicles, unusual vehicle types, snowy/icy roads and higher-speed roads were among the circumstances overrepresented in crashes involving AEB-equipped striking vehicles. The finding motivates broader testing. It does not imply AEB causes every crash in those circumstances or gives an individual Tesla a numerical failure probability. [5]

Engineering model

A system must act on incomplete observations while the scene changes. Triggering too readily can create unwanted interventions; triggering too late can leave less distance to respond. This is a general detection-and-control trade-off, not a claim about Tesla’s internal confidence thresholds. A chart of visible camera coverage would not solve it: seeing an object does not establish its future path or whether an intervention is appropriate.

Explanation

The Model Y camera chapter identifies obstructions and visibility conditions as performance concerns. Clean lenses and calibration address prerequisites, not every ambiguity. Separate sensing from stopping physics: better braking cannot compensate for every missed observation. [7]

Explanation

A useful incident record includes the vehicle, installed build, feature settings, conditions, displayed message and available unedited footage. Avoid guessing whether a visible slowing was a driver input or an automated action. Without the relevant evidence, an honest account can describe the observation while leaving the mechanism unresolved.

Explanation

The label Collision Evasion should be tied to the exact software release and its own description. It must not inherit every property of FCW, AEB or supervised driving merely because those functions share a vehicle. Likewise, the absence of that label in one public manual is not proof that no release contains additional behaviour. The correct question is which documented behaviour applies to which configuration.

Explanation

For an automatic-engagement claim, look for the release identifier, supported hardware, market, prerequisites, operating conditions and instructions for driver control. A social clip can be an observation; a complete primary release note is stronger evidence for the intended feature conditions. This review did not establish an accessible exact primary release note covering every re-engagement claim, so the guide does not turn secondary descriptions into operating instructions.

Explanation

Use the linked software tracker for current reports, then confirm the installed car’s own release notes and manual. A comprehensive explanation should make uncertainty specific: what is documented, what is calculated and what remains unverified. It should neither promise a rescue manoeuvre nor dismiss a reported feature simply because the name resembles an older safety function.