Fixing Accessibility Gaps in Google Maps Routing

by

Summary

The user, Christa, identified a functional gap in Google Maps regarding accessibility modeling. Standard routing engines prioritize vehicles (cars) or pedestrians (walking). For users with electric wheelchairs—which operate at an intermediate velocity (6 km/h)—there is no dedicated profile. This results in inaccurate travel time estimations, as users are forced to use “walking” profiles that do not account for the specific constant velocity constraints of assistive mobility devices.

Root Cause

The failure stems from a lack of specialized entity modeling within the routing engine’s cost function.

  • Discrete Profile Limitation: Routing engines typically use a finite set of profiles (Walking, Cycling, Driving).
  • Velocity Mismatch: A walking profile assumes a variable human pace (often 4-5 km/h with stops/starts), whereas an electric wheelchair maintains a steady-state mechanical velocity.
  • Edge Property Neglect: The engine treats all “pedestrian” paths as equal, failing to distinguish between paths optimized for mobility aids versus those optimized for standard pedestrian foot traffic.

Why This Happens in Real Systems

In large-scale production systems, this is a classic case of Optimization Bias.

  • The 80/20 Rule: Engineering teams prioritize the 80% use case (car drivers and walkers) to minimize complexity.
  • Model Sparsity: Adding a new “profile” isn’t just changing a number; it requires defining new rules for turn penalties, slope gradients, and surface types that specifically affect electric motors.
  • Data Siloing: Accessibility data (wheelchair ramps, sidewalk widths) is often stored in separate layers from the core topological graph, making real-time multi-modal routing computationally expensive.

Real-World Impact

When a system fails to model a specific user segment, the consequences move from “inconvenience” to safety and autonomy risks:

  • Planning Fallacy: Users provide incorrect arrival times for appointments, leading to social and medical friction.
  • Energy Management Failure: For electric users, knowing the exact distance is critical for battery life management. An incorrect estimate can lead to a complete power failure mid-journey.
  • Inclusion Gap: The software effectively “erases” a demographic by providing tools that are technically functional but practically useless.

Example or Code (if necessary and relevant)

class RoutingEngine:
    def __init__(self):
        self.profiles = {
            "walking": {"avg_speed_kmh": 4.5, "stops_per_km": 2},
            "cycling": {"avg_speed_kmh": 15.0, "stops_per_km": 0.5},
            "driving": {"avg_speed_kmh": 50.0, "stops_per_km": 0.1}
        }

    def calculate_travel_time(self, distance_km, profile_name):
        if profile_name not in self.profiles:
            raise ValueError("Profile not supported")

        profile = self.profiles[profile_name]
        # A simple model that ignores the nuances of assistive tech
        return distance_km / profile["avg_speed_kmh"]

# The Problem: The user's specific profile is missing
engine = RoutingEngine()

# Christa's actual speed is 6 km/h
# She is forced to use 'walking'
estimated_time_walking = engine.calculate_travel_time(10, "walking")
print(f"Walking Estimate: {estimated_time_walking:.2f} hours")

# The desired behavior (The Fix)
# engine.add_profile("electric_wheelchair", {"avg_speed_kmh": 6.0})

How Senior Engineers Fix It

A Senior Engineer looks beyond the immediate “missing feature” and addresses the system architecture:

  • Extensible Profile Registry: Implement a plugin-based architecture where new movement profiles (like “Electric Wheelchair”) can be injected without modifying the core Dijkstra or A* algorithm.
  • Parameterization of Edge Weights: Instead of a single speed, use multi-dimensional weight vectors (slope, surface, width) that interact with the specific velocity and torque profiles of the user’s device.
  • LLM-Assisted Parameter Extraction: To address Christa’s point about AI, use Large Language Models as a natural language interface to map user descriptions (“I have a 6 km/h chair”) directly into the specific telemetry parameters required by the routing engine.

Why Juniors Miss It

  • Focus on Logic, Not Domain: Juniors often focus on making the “walking” math work perfectly, rather than questioning if “walking” is the correct abstraction for the user.
  • The “Happy Path” Bias: They design for the average user and assume that “close enough” (using walking instead of wheelchair) is an acceptable approximation.
  • Feature vs. System Thinking: A junior sees this as “add a 6 km/h button”; a senior sees this as “the routing engine needs a more granular way to model specialized mobility profiles.”

Leave a Comment