Real self-driving stacks (Waymo, Cruise, Tesla FSD) never trust one sensor alone. LIDAR gives precise 3D range but struggles in fog/rain; cameras give rich semantic detail but degrade with glare and poor visibility; radar is weather-robust but coarse. Each active sensor here computes an independent per-object detection probability from distance and range, then the perception stack fuses them probabilistically — the more independent sensors agree, the higher the combined confidence, even if any single sensor is unreliable.
effRange_i = baseRange_i * (1 - fog * weatherLoss_i)
p_i(d) = reliability_i * clamp(1 - d/effRange_i, 0, 1) if d < effRange_i, else 0
P_fused = 1 - Π (1 - p_i) over all active sensors i
stopping_distance = v²/(2·μ·g) + v·t_reaction
μ = 0.7 (dry asphalt), g = 9.81 m/s², t_reaction = 1.5 s
- LIDAR / Camera / Radar buttons — toggle each sensor on/off and watch the fused confidence and cone visuals change.
- Fog / weather — cuts LIDAR and camera range sharply (they scatter off water droplets); radar is barely affected.
- Ego speed — feeds directly into the physical stopping-distance formula above.
- Traffic density — number of tracked vehicles/pedestrians the fusion stack must classify simultaneously.
This is why regulators (NHTSA, UK's CCAV) require redundant, complementary sensor suites before certifying any SAE Level 4+ system for public roads.