How Map Priors and Traffic Rules Reduce Risk at Occluded Intersections

When part of an intersection is occluded, the vehicle’s sensors alone give an incomplete picture. Map priors and traffic-rule signals constrain what the system treats as plausible in that hidden region, turning an unbounded uncertainty problem into a structured risk assessment that a planner can act on.

What map priors provide

High-definition (HD) and vector maps encode fixed, lane-level structure and semantics that are stable over short time horizons. Useful priors include:

– lane centerlines and lane connectivity (legal approach/egress directions),

– crosswalk and sidewalk locations, curb edges, and pedestrian-only zones,

– speed limits, turn restrictions, and intersection topology (stop signs, yield lines),

– typical clearance envelopes (building footprints, parked-vehicle zones) and probable occluder locations.

These priors let the perception and prediction stack assign higher probability to actors emerging from sidewalks or legal lanes and near crosswalks, and much lower probability to implausible trajectories (e.g., appearing from inside a building).

How traffic rules shape conservative behavior

Explicit rule encodings (stop/yield obligations, right-of-way, speed limits, turning bans) translate map semantics into decision constraints a planner can enforce:

– Hard constraints: require a full stop or yielding before entering sight-limited areas.

– Soft costs: increase the penalty for entering a crosswalk or proceeding at high speed when occlusion is present, biasing the optimizer toward slower, shorter gaps.

– Legality filters: disallow trajectories that would violate lane-level turn prohibitions even if they appear expedient.

Combining priors, rules, and sensor evidence

Practical systems fuse three information streams into a single risk model:

1) Map-derived priors define where actors are likely to originate and which headings are legal.

2) Traffic-rule logic imposes mandatory behaviour (stop/yield) and shapes the planner’s cost function.

3) Live sensor cues (lead-vehicle braking, wheel orientation of off-camera cars, sound, partial detections) update posterior probabilities for hidden actors.

Fusing these yields calibrated occupancy hypotheses for occluded cells: for example, a high prior for a pedestrian at a sidewalk corner plus a sudden brake by the car ahead should raise the pedestrian-existence probability substantially and trigger a defensive deceleration.

Implementation patterns for product teams

– Represent occluded space explicitly (grid or vector cells) and seed it with map-based prior probabilities by semantic class (pedestrian, bicycle, car).

– Encode traffic rules as both hard constraints (safety-critical) and tunable costs (comfort/performance trade-offs) so behaviour can be audited and tuned separately.

– Use hierarchical planners: a fast reflex layer enforces rules and hard safety margins; a mid-level planner reasons over map-informed priors to pick conservative maneuvers; a trajectory optimizer refines comfort while respecting constraints.

– Update priors online from fleet telemetry or local observations (e.g., common pedestrian emergence points) to reduce false positives without compromising safety.

– Instrument logging and scenario-driven testing (edge-case catalogs that combine occlusion + rule intersections) to validate that priors and rules produce the intended conservative behaviours.

Common failure modes and mitigations

– Outdated maps: fall back to conservative default priors (assume pedestrian presence) and increase stopping margins when map confidence is low.

– Overconfident priors: clamp prior probabilities to avoid dismissing rare-but-dangerous events; keep a minimum hazard weight for any occluded walkable area.

– Conflicting signals (e.g., green light but stop sign for turning lane): resolve via an explicit rule hierarchy that prioritizes safety-critical requirements.

Bottom line

Map priors and traffic rules do not remove uncertainty; they structure it. By restricting where actors can legally and plausibly appear, and by converting regulations into auditable constraints and costs, systems reduce false negatives and choose conservative, explainable behaviours at occluded intersections—improving both safety and predictability.

Sources

• Driving by the Rules: A Benchmark for Integrating Traffic Sign Regulations into Vectorized HD Map (MapDR) (Xinyuan Chang et al. (project page); 2025-10-01)
• A Survey of World Models for Autonomous Driving (ArXiv (survey); 2025-09-07)