Collision Avoidance Systems

Last modified: Jul 29, 2026

Collision avoidance is a family of warning and intervention functions, not one feature with a universal sensor set or performance level.

Sensors used

Collision-avoidance functions can use different combinations of the modalities described in Sensors and Cameras. The appropriate set depends on the target, direction, speed, operating conditions and whether the vehicle warns, brakes or steers.

Sensor count is not a safety rating. Sensor Fusion covers complementarity and redundancy, Calibration, Cleaning and Sensor Health covers degraded sensing, and How to Validate Sensor and Perception Claims explains why complete vehicle behavior must be tested.

The safety chain

A collision-avoidance function usually performs four tasks:

  1. perceive vehicles, vulnerable road users, boundaries or free space;
  2. predict how their paths may intersect;
  3. decide whether and when to warn or intervene;
  4. request braking, steering or both while preserving stability.

Errors can enter at every stage. A target may be hidden, classified incorrectly, move unexpectedly or leave too little grip for the commanded maneuver.

Warning, support and intervention

Related names describe different actions:

  • forward collision warning alerts the driver;
  • brake support increases braking when the driver's input is insufficient;
  • automatic emergency braking can apply the brakes without driver input;
  • automatic emergency steering changes the path;
  • blind-spot intervention responds to an unsafe lane change;
  • rear automatic braking addresses selected reversing conflicts.

Momentary warnings and emergency interventions remain SAE Level 0 because they do not provide sustained control. They can operate on a vehicle that has no Level 1 or Level 2 convenience assistance.

Scenario coverage

Capability should be stated by scenario:

  • target type: car, truck, motorcycle, cyclist, pedestrian or object;
  • direction: forward, crossing, oncoming, turning or reversing;
  • speed and relative speed;
  • daylight, darkness and weather;
  • straight road, curve, junction or lane change;
  • warning, impact mitigation or full avoidance.

"AEB with pedestrian detection" does not establish cyclist, motorcycle, junction or high-speed coverage.

Sensor and software diversity

Cameras provide classification and visual context, radar measures range and relative velocity, ultrasonic sensors cover the near field, and lidar can provide detailed depth. A capable architecture can use several sensors, but redundancy depends on independent failure modes and software design, not sensor count alone.

The perception problem is explored in Sensors and Perception, while Software Architecture and AI explains planning, control and validation.

Regulation and ratings

The EU General Safety Regulation requires several collision-avoidance technologies under its implementation timetable. The United States has adopted FMVSS No. 127 for automatic emergency braking on new passenger vehicles, with requirements taking effect according to the federal timetable.

Legal tests establish minimum defined performance. Consumer programs such as Euro NCAP cover broader scenarios and update protocols as crash patterns and technology evolve. Neither guarantees avoidance in every real-world event.

Sources

More information