On Assumptions with Respect to Occlusions in Urban Environments for Automated Vehicle Speed Decisions

TL;DR

This paper analyzes how occlusions in urban environments influence speed decisions for automated vehicles. It introduces a dynamic speed-limit model derived from a pedestrian potentially emerging from occlusions, incorporating reaction time and braking constraints, and evaluates it with example assumptions in simulation. The results show that explicit occlusion treatment can cause substantial speed reductions and may not eliminate residual risk, highlighting the need for explicit disclosure of assumptions and broader societal discussion to determine acceptable levels of risk. The work argues that regulatory guidance, infrastructure adaptation, and public consensus are essential to meaningfully reduce pedestrian risk without degrading traffic flow, and frames occlusion handling as a key component of the safety argument for automated driving systems.

Abstract

Automated driving systems are subject to various kinds of uncertainty during design, development, and operation. These kinds of uncertainty lead to an inherent risk of the technology that can be mitigated, but never fully eliminated. Situations involving obscured traffic participants have become popular examples in the field to illustrate a subset of these uncertainties that developers must deal with during system design and implementation. In this paper, we describe necessary assumptions for a speed choice in a situation in which an ego-vehicle passes parked vehicles that generate occluded areas where a human intending to cross the road could be obscured. We develop a calculation formula for a dynamic speed limit that mitigates the collision risk in this situation, and investigate the resulting speed profiles in simulation based on example assumptions. This paper has two main results: First, we show that even without worst-case assumptions, dramatically reduced speeds would be driven to avoid collisions. Second, we highlight that design decisions regarding occlusion treatment are directly related to the risk that automated vehicles pose to pedestrians in urban environments. In this respect, we conclude that there needs to be a broader discussion about acceptable assumptions.

Figures (6)

• Figure 1: Scenario examined of an automated vehicle passing a large parked vehicle
• Figure 2: Ego-vehicle passes a parked car $m$ while following path $\mathrm{\Gamma}_\mathrm{Ego}$. From the ego-vehicle's perspective, there is an occluded area between parked cars $m$ and $m+1$ (indicated in the graphic). At $\vec{p}_{\mathrm{O}m}$ an obscured relevant object could be present.
• Figure 3: Simplified relations when passing a large vehicle parked next to the lane on a straight route
• Figure 4: Speed limit profile of the ego-vehicle while passing a large parked vehicle with a lateral distance of 1 with the introduced example assumptions (solid) and extremely pessimistic assumptions (dashed). The highlighted area indicates the position of the parked vehicle next to the ego-lane and the dotted gray line indicates the posted speed limit.
• Figure 5: Simulated scenario of the UNICARagil concept vehicle passing three critical occlusions in a narrow street