An Empirical Analysis of Cooperative Perception for Occlusion Risk Mitigation

Aihong Wang; Tenghui Xie; Fuxi Wen; Jun Li

cs.ROFeb 26, 2026

An Empirical Analysis of Cooperative Perception for Occlusion Risk Mitigation

Authors: Aihong Wang, Tenghui Xie, Fuxi Wen, Jun Li

Organizations: School of Vehicle and Mobility, Tsinghua University, China.

Abstract

Occlusions present a significant challenge for connected and automated vehicles, as they can obscure critical road users from perception systems. Traditional risk metrics often fail to capture the cumulative nature of these threats over time adequately. In this paper, we propose a novel and universal risk assessment metric, the Risk of Tracking Loss (RTL), which aggregates instantaneous risk intensity throughout occluded periods. This provides a holistic risk profile that encompasses both high-intensity, short-term threats and prolonged exposure. Utilizing diverse and high-fidelity real-world datasets, a large-scale statistical analysis is conducted to characterize occlusion risk and validate the effectiveness of the proposed metric. The metric is applied to evaluate different vehicle-to-everything (V2X) deployment strategies. Our study shows that full V2X penetration theoretically eliminates this risk, the reduction is highly nonlinear; a substantial statistical benefit requires a high penetration threshold of 75-90%. To overcome this limitation, we propose a novel asymmetric communication framework that allows even non-connected vehicles to receive warnings. Experimental results demonstrate that this paradigm achieves better risk mitigation performance. We found that our approach at 25% penetration outperforms the traditional symmetric model at 75%, and benefits saturate at only 50% penetration. This work provides a crucial risk assessment metric and a cost-effective, strategic roadmap for accelerating the safety benefits of V2X deployment.

Figures & tables

Panel of Figure 1 — Fig. 1: Illustrative examples of critical sensory blind spots for a vehicle, with the resulting occluded regions circled in red. (a) A static occlusion caused by the A-pillar, which obscures a child about to retrieve a ball. (b) A dynamic occlusion caused by a large truck, which conceals a pedestrian running across the road from its front.

Symbol	Description
Risk Assessment Framework
$RU_{i},RU_{j}$	Subject road user and interacting road user
$t,T$	Discrete time frame index and total time horizon
$f_{i,j}(t)$	Instantaneous risk intensity imposed by the interacting road user
$A_{k}$	Risk integral of a continuous occlusion event
$F_{i,j}$	Maximal sustained risk among all occlusion events

TABLE I: Summary of Key Notations

Panel of Figure 3 — Fig. 2: The RTL computational pipeline illustrating the hierarchical flow from trajectory inputs to system-level safety evaluation.

Panel of Figure 4 — Fig. 3: Demonstration of two risk profiles captured by the proposed metric.

Panel of Figure 5 — Fig. 4: Decision logic for the coefficient $k_{\rm dynamic}$ in dynamic scenarios.

Panel of Figure 6 — Fig. 5: Visualization of spatial risk aggregation heatmaps for scenarios from the SIND and Waymo datasets. The heatmaps are generated by accumulating RTL values of high-risk events at their peak-risk locations. Gray areas denote roads, and the color gradient indicates risk intensity, with warmer colors indicating higher accumulated RTL. The first row (a-d) displays intersection scenarios from the SIND dataset, comparing Veh-Veh and Veh-VRU at the Tianjin and Changchun intersections. The second row (e-h) presents various Veh-Veh interaction scenarios from the Waymo dataset, including a T-junction, a multi-fork road, a merging lane, and an ordinary road.

Panel of Figure 7 — Fig. 6: Correlation between RTL and traffic exposure metrics. Each subplot jointly visualizes distributional risk via box plots on the left Y-axis against a corresponding traffic exposure metric, such as Traffic Volume or VRU Ratio, via bar charts on the right Y-axis. The figure presents analyses for (a-b) Veh-Veh and Veh-VRU in the SIND dataset and (c) Veh-Veh in the Waymo dataset.

Panel of Figure 8 — Fig. 7: Representative risk events identified by the proposed metric across both datasets. In each subfigure, the circled pair highlights a risk interaction, with the circle’s color indicating the single-frame risk value at the illustrated instant. (a) A non-motorized vehicle running a red light at an intersection. (b) A VRU within a vehicle’s blind spot during a turning maneuver at a road edge convergence zone. (c) A bus occupies a non-motorized lane, forcing a VRU into the traffic flow. (d) A conflict during an unprotected left turn. (e) A prolonged car-following scenario exhibiting high cumulative risk. (f) A conflict arising from a lane merge.

Dataset	Scenario	$C_{QD}$		$CV_{MAD}$ (%)
Dataset	Scenario	MTL	RTL	MTL	RTL
SIND	Tianjin (Veh-Veh)	0.5722	0.5897	38.32	59.45
	Tianjin (Veh-VRU)	0.3701	0.5160	37.76	58.31
	Changchun (Veh-Veh)	0.5232	0.7514	47.96	72.33
	Changchun (Veh-VRU)	0.2941	0.7259	27.81	65.99
	SIND Mean	0.4399	0.6458	37.96	64.02

TABLE II: Comparison of

C_{QD}

and

CV_{MAD}

Between MTL and RTL

Scenario	Normalized Top 10% Mean RTL at Penetration Rate (%)
Scenario	25%	50%	75%	90%	100%
Changchun (Veh-Veh)	91.85	81.78	63.73	42.24	1.24
Tianjin (Veh-Veh)	85.49	64.57	36.83	22.40	0.00
Xi’an (Veh-Veh)	87.35	75.19	57.21	32.30	0.37
Chongqing (Veh-Veh)	88.03	71.70	43.26	25.43	0.00
Changchun (Veh-VRU)	91.61	81.22	62.64	42.02	0.02

TABLE III: Quantitative Comparison of Normalized RTL Reduction Across V2X Penetration Rates

Parameter	Top 10% Mean RTL at Penetration Rate (ms)
Parameter	0%	25%	50%	75%	90%	100%
Conventional Paradigm
Baseline	124.33	103.77	59.48	42.60	15.78	0.00
K1 ( $k_{\mathrm{overlap}}\downarrow$ )	76.67	63.18	41.04	27.15	17.15	0.00
K2 ( $k_{\mathrm{overlap}}\uparrow$ )	203.47	162.32	102.86	69.51	36.35	0.00
K3 ( $k_{\mathrm{no\mbox{-}overlap}}\downarrow$ )	112.29	81.71	60.48	29.22	27.10	0.00

TABLE IV: Parameter Sensitivity Analysis: RTL Comparison Between Paradigms

Autonomous Driving Safety Evaluation

Autonomous driving safety evaluation studies how to measure, test, and improve the safety of self-driving systems in routine and hazardous conditions.

See all 158 papers

Learning A Unified Risk Map for Autonomous Driving in Partially Observable Environments
May 21, 2026Jie Jia, Yaofeng Su, Zeyu Bao +4Occlusion Reasoning Autonomous Driving Perception
Fudan University, China · Tongji University, China
CooperDrive: Enhancing Driving Decisions Through Cooperative Perception
Apr 15, 2026Deyuan Qu, Qi Chen, Takayuki Shimizu +1Cooperative Perception Communication-Efficient Collaborative Perception
Toyota InfoTech Labs, Mountain View, CA, USA
Occlusion-Aware Contingency Safety-Critical Planning for Autonomous Driving
Feb 10, 2025Lei Zheng, Rui Yang, Minzhe Zheng +3Safe Motion Planning Motion Planning
Robotics and Autonomous Systems Thrust, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou 511453, China · School of Engineering, Great Bay University, Dongguan 523000, China · Cheng Kar-Shun Robotics Institute, The Hong Kong University of Science and Technology, Hong Kong SAR, China

May 21, 2026J/K move · Enter open · S save

Learning A Unified Risk Map for Autonomous Driving in Partially Observable Environments

Jie Jia, Yaofeng Su, Zeyu Bao, Yun Hong+3

Fudan University, China · Tongji University, China

Occlusion Reasoning Autonomous Driving Perception Autonomous Driving Planning

Open paper AskPDF arXiv

Occlusion-aware prediction remains a critical challenge in autonomous driving due to the inherent uncertainty of unobserved regions. Existing approaches either overestimate risk based on reachable states or struggle to predict accurate trajectories under high occlusion uncertainty. To address these limitations, we propose a unified risk map modeling and learning framework for partially observable environments. Our method integrates traffic flow risk and collision risk through spatiotemporal modeling, enabling fine-grained assessment of occlusion-induced hazards. To address the scarcity of scenarios involving occluded interactions, we introduce a diffusion-based scenario generation framework that produces realistic yet adversarial scenarios. We integrate the modeling and learning of a unified risk map into a framework that supports risk-aware planning under partial observability. Experiments on the Waymo Open Motion Dataset show that our method significantly outperforms the state-of-the-art occlusion-aware baseline, improving minimum time-to-collision by 0.78 times and average time-to-collision by 1.67 times. The proposed framework offers a comprehensive and practical solution for risk-aware planning in partially observable environments.

Abstract

Autonomous Driving Safety Evaluation

Explore similar work

Learning A Unified Risk Map for Autonomous Driving in Partially Observable Environments

CooperDrive: Enhancing Driving Decisions Through Cooperative Perception

Occlusion-Aware Contingency Safety-Critical Planning for Autonomous Driving