Risk-Aware Occupancy for Safety-Oriented End-to-End Autonomous Driving

Jiaxing Chen; Hengduo Zou; Yiren Zhao; Bolin Gao

cs.AISep 18, 2026

Risk-Aware Occupancy for Safety-Oriented End-to-End Autonomous Driving

Authors: Jiaxing Chen, Hengduo Zou, Yiren Zhao, Bolin Gao

Organizations: School of Vehicle and Mobility, Tsinghua University · The Hong Kong University of Science and Technology (Guangzhou)

Abstract

Conventional end-to-end driving systems model the environment with sparse objects and lane elements. While efficient, this paradigm discards planning-critical information in crowded and occluded scenarios, particularly for unstructured obstacles, ambiguous free space, and complex interactions. We propose risk-aware occupancy, a dense BEV representation that explicitly fuses geometric occupancy, map-derived traffic constraints, and future dynamic-agent occupancy as complementary risk signals. Built upon this representation, we develop ROIDrive, an instance-centric end-to-end framework with a dedicated risk-aware occupancy branch. The predicted occupancy is tokenized via sliding-window sampling and injected into planning queries via cross-attention, while temporal query consistency mitigates unreliable flickering queries. We also contribute RiskOcc4D-nuScenes, a benchmark derived from nuScenes and Occ3D-nuScenes with four automated annotation pipelines for multi-dimensional risk supervision. Experiments on representative occupancy architectures verify the learnability and transferability of our representation. Integrated with GenAD, it reduces collision rates by 35.0% (UniAD metric) and 52.9% (ST-P3 metric), confirming the efficacy of the proposed representation modality.

Figures & tables

Figure 1: Motivation and design of risk-aware occupancy. Left: Limitations of sparse agent/map detection and naive 3D occupancy, and human-inspired requirements for dense representations covering geometry, drivability, traffic rules and vulnerable road users. Middle: Comparison between conventional decoupled E2E pipelines and our unified prediction-planning framework, which constructs global and dynamic risk-aware occupancy maps by encoding collision and traffic-regulation risks. Right: Collision-rate benchmarks on UniAD and ST-P3 metrics, verifying the safety improvement of the proposed risk-aware occupancy representation.

Figure 2: ROIDrive architecture. A BEVFormer backbone yields the BEV feature $B$ ; the risk-aware occupancy (RiskOcc) branch regresses a normalized risk-aware occupancy via a Bi-Mamba2 block and a class-to-risk lookup table (optionally fused with an SDMap+ prior), tokenizes it by sliding-window sampling, and injects it into the queries via cross-attention. Temporal Query Consistency (TQC) then gates the queries by cross-frame reliability before a VAE generative planner decodes multimodal trajectories.

Figure 3: The four automated annotation pipelines building RiskOcc4D-nuScenes, fusing static, regulatory, drivable-area, and dynamic-future cues into the global and dynamic risk-aware occupancy maps.

Baselines	Release	Metrics
		Global					Dynamic	Infer. Speed
		mIoU $\uparrow$	MAE $\downarrow$	MAE@15 $\downarrow$	MAE@30 $\downarrow$	SSIM $\uparrow$	Mask-MAE $\downarrow$	FPS $\uparrow$
PanoOcc-S ( Wang et al. 2024 )	CVPR 24	14.54	0.121	0.022	0.079	0.131	0.732	6.0
PanoOcc ( Wang et al. 2024 )	CVPR 24	14.71	0.121	0.022	0.079	0.131	0.731	1.9
OccProphet ( Chen et al. 2025 )	ICLR 25	17.39	0.099	0.017	0.067	0.355	0.524	0.7
ViewFormerOcc ( Li et al. 2024 )	ECCV 24	21.38	0.093	0.015	0.060	0.379	0.734	19.0

Table 1: Feasibility benchmark for predicting risk-aware occupancy with representative 3D occupancy backbones. Best values are shown in bold .

Figure 4: Architecture of the RiskOcc-GenAD fusion experiment: the predicted risk-aware occupancy map is sampled by a sliding window and fused with GenAD’s instance tokens via cross-attention, entering as an additional input while GenAD’s protocols stay unchanged.

Methods	ST-P3 Metrics $\downarrow$				UniAD Metrics $\downarrow$
Methods	1 s	2 s	3 s	Avg.	1 s	2 s	3 s	Avg.
ST-P3	0.23	0.62	1.27	0.71	-	-	-	-
OccNet	0.21	0.59	1.37	0.72	-	-	-	-
VAD-Tiny	0.21	0.35	0.58	0.38	-	-	-	-
VAD-Base	0.07	0.17	0.41	0.22	0.01	0.30	0.95	0.42
FusionAD	0.25	0.13	0.25	0.21	-	-	-	-

Table 2: Collision rate (%) comparison among E2E methods. Ours = GenAD ( Zheng et al. 2024 )

+

RiskOcc Map; P. = progressive training. Best values are shown in bold .

Inputs	ST-P3 Metrics $\downarrow$				UniAD Metrics $\downarrow$
Inputs	1 s	2 s	3 s	Avg.	1 s	2 s	3 s	Avg.
RiskOcc Map	0.02	0.04	0.19	0.08	0.00	0.14	0.64	0.26
3D-Occ	0.09	0.18	0.33	0.20	0.08	0.29	0.74	0.37
Binary Map	0.11	0.21	0.39	0.24	0.14	0.39	0.90	0.48
Gaussian Map	0.21	0.32	0.37	0.33	0.27	0.51	0.86	0.55

Table 3: Ablation of intermediate representations on collision rate (%). Best in bold .

Methods			UniAD Metrics								ST-P3 Metrics
M1	M2	M3	L2 ( m ) $\downarrow$				Collision Rate (%) $\downarrow$				L2 ( m ) $\downarrow$				Collision Rate (%) $\downarrow$
M1	M2	M3	1 s	2 s	3 s	Avg.	1 s	2 s	3 s	Avg.	1 s	2 s	3 s	Avg.	1 s	2 s	3 s	Avg.
✓			0.38	0.85	1.52	0.92	0.07	0.07	0.64	0.26	0.29	0.51	0.78	0.52	0.08	0.07	0.21	0.12
✓	✓		0.21	0.58	1.22	0.67	0.08	0.16	0.84	0.36	0.16	0.31	0.56	0.34	0.08	0.10	0.27	0.15
✓	✓	✓	0.21	0.59	1.26	0.69	0.00	0.14	0.86	0.33	0.16	0.32	0.57	0.35	0.00	0.04	0.21	0.08

Table 4: ROIDrive framework ablation. M1: risk-aware occupancy perception; M2: ego-token prior initialization; M3: Temporal Query Consistency (TQC). Best values are shown in bold .

Components		UniAD Collision Rate $\downarrow$				ST-P3 Collision Rate $\downarrow$
C1	C2	1 s	2 s	3 s	Avg.	1 s	2 s	3 s	Avg.
✓		0.08	0.18	0.84	0.37	0.08	0.11	0.30	0.16
	✓	0.06	0.27	0.54	0.29	0.05	0.12	0.22	0.13
✓	✓	0.07	0.07	0.64	0.26	0.08	0.07	0.21	0.12

Table 5: RiskOcc-branch component ablation without ego-token initialization or TQC. C1: Bi-Mamba2; C2: SDMap+ fusion. Best values are shown in bold .

End-to-End Autonomous Driving

End-to-end autonomous driving learns to turn sensor observations and driving context into vehicle actions or trajectories.

See all 227 papers

Driving on Registers, Reasoning on Risk: Risk-Aware Occupancy for Register-Based End-to-End Autonomous Driving
Sep 18, 2026Jiaxing Chen, Hengduo Zou, YuKai Qin +3End-to-End Autonomous Driving Autonomous Driving Planning
School of Vehicle and Mobility, Tsinghua University · The Hong Kong University of Science and Technology (Guangzhou) · Neolix, Beijing, China
Data-Driven Risk Fields for Safer End-to-End Autonomous Driving
Sep 9, 2026Yuanxin Tian, Zhiyuan Liu, Jinhao Li +9End-to-End Autonomous Driving Safe Motion Planning
Tsinghua University · Yinwang · KAIST
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

Sep 18, 2026J/K move · Enter open · S save

Driving on Registers, Reasoning on Risk: Risk-Aware Occupancy for Register-Based End-to-End Autonomous Driving

Jiaxing Chen, Hengduo Zou, YuKai Qin, Yiren Zhao+2

School of Vehicle and Mobility, Tsinghua University · The Hong Kong University of Science and Technology (Guangzhou) · Neolix, Beijing, China

End-to-End Autonomous Driving Autonomous Driving Planning Multimodal Trajectory Prediction 3D Occupancy Prediction Autonomous Driving Datasets

Open paper AskPDF arXiv

Multimodal trajectory prediction improves behavioral coverage in end-to-end autonomous driving, but existing methods remain limited by sparse scene representations. Incomplete evidence leads to low-quality candidate generation and unreliable ranking among geometrically similar trajectories. On a register-based baseline, bad and poor candidates constitute 19.74% of the candidate set, while the oracle-best candidate ranks only 33.9th on average. We propose RRDrive, which introduces risk-aware occupancy as a dense, temporally aligned, and trajectory-queryable representation. Its global structure guides high-quality multimodal generation, while candidate-conditioned risk queries support fine-grained selection. We further construct RiskOcc4D-NAVSIM with automatic risk annotations. RRDrive achieves a selected-trajectory PDMS of 0.951, representing a 1.5% relative improvement over the baseline (0.937), and improves the average candidate PDMS by 7.7%. In challenging scenes, it improves candidate PDMS by 30.2% and increases the Spearman correlation among good candidates by 0.41, from 0.26 to 0.67. To move beyond this oracle setting, we further develop an external RiskOcc predictor, a perception module that estimates risk-aware occupancy directly from sensor inputs. The competitive performance validates the representation's feasibility.

Abstract

End-to-End Autonomous Driving

Explore similar work

Driving on Registers, Reasoning on Risk: Risk-Aware Occupancy for Register-Based End-to-End Autonomous Driving

Data-Driven Risk Fields for Safer End-to-End Autonomous Driving

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