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; Lidong Yu; Bolin Gao

cs.AISep 18, 2026

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

Authors: Jiaxing Chen, Hengduo Zou, YuKai Qin, Yiren Zhao, Lidong Yu, Bolin Gao

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

Abstract

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.

Figures & tables

Figure 1: Motivation and overall performance of RRDrive. Register-based baselines face both candidate generation waste and ranking errors in complex scenes, driven by unmodeled local future risks. Left: Multi-view inputs and temporal evolution of risk-aware occupancy. Right: Trajectory comparison and quantitative improvements validate our dense trajectory-queryable risk representation.

Figure 2: Overall architecture of RRDrive. A risk tokenizer encodes BEV risk-aware occupancy, which is fused with camera scene tokens for risk-conditioned candidate generation and read out along each candidate trajectory for risk-aware selection.

Figure 3: RiskOcc-conditioned candidate generation module. Risk tokens and camera scene tokens are combined by bidirectional cross-attention with LayerScale gating, and the fused context conditions the trajectory decoder so that the finite candidate budget is steered toward feasible, high-quality behaviors.

Figure 4: Risk-aware candidate ranking module. For each candidate, the ego footprint samples the time-aligned risk field along the trajectory to form a local risk profile, which augments the risk-sensitive scorer heads and sharpens ranking among geometrically similar good candidates.

Baselines & Oracle
Methods	PDMS $\uparrow$	NC $\uparrow$	DAC $\uparrow$	EP $\uparrow$	TTC $\uparrow$	Comf. $\uparrow$
Human Driver 2024	94.8	100.0	100.0	87.5	100.0	99.9
PDMS-Closed 2023	89.1	94.6	99.8	99.9	86.9	89.9
Ego-stat. MLP 2024	65.6	93.0	77.3	62.8	83.6	100.0
VLA-based Methods
UniVLA 2025	81.7	96.9	91.1	76.8	91.7	96.7

Table 1: Comparison of results on the NAVSIM benchmark. All metrics are higher-is-better. Best results are highlighted in bold. VLA-based and E2E methods are sorted ascendingly by PDMS.

\uparrow

\downarrow

: higher / lower is better.

Figure 5: Qualitative comparison of candidate generation on two generation-challenging scenes. Compared with DrivoR (third column), RRDrive (fourth column) concentrates the finite proposal budget within feasible, high-quality regions aligned with the expert trajectory, producing more good candidates and higher mean candidate PDMS.

Variant	Module	NC $\uparrow$	DAC $\uparrow$	DDC $\uparrow$	TTC $\uparrow$	EP $\uparrow$	Comf. $\uparrow$	PDMS $\uparrow$	$\Delta$
Baseline (no RiskOcc)	–	99.02	98.92	97.23	96.72	89.91	100.00	93.67	–
+ global RiskOcc encoder	Gen.	99.24	99.08	97.51	97.34	90.96	99.99	94.49	$+0.82$
+ trajectory-conditioned sampling	Rank.	99.30	99.02	97.56	97.57	91.25	99.99	94.68	$+0.19$
+ bidirectional cross-modal fusion	Gen.	99.37	99.04	97.46	97.78	91.51	100.00	94.90	$+0.22$
+ TTC/EP & RiskOcc refine	Rank.	99.34	98.98	97.41	97.55	92.19	100.00	95.06	$+0.16$

Table 2: Cumulative ablation on navtest (argmax, no re-ranking), decomposed into all PDMS sub-metrics. Module tags the primary pathway each component serves: Gen. = candidate generation, Rank. = candidate ranking. All metrics are higher-is-better; NC = no at-fault collision, DAC = drivable-area compliance, DDC = driving-direction compliance, TTC = time-to-collision, EP = ego-progress, Comf. = comfort.

\Delta

is the absolute PDMS gain over the preceding row. Rows are model-lineage proxies: each step carries minor co-changes, so tags denote the dominant pathway rather than an exclusive one.

Metric	Baseline	Ours	$\Delta(\%)$
Good ratio $\uparrow$	80.3%	88.7%	$+10.5$
Bad ratio $\downarrow$	13.6%	7.6%	$-44.1$
Mean cand. PDMS $\uparrow$	0.797	0.858	$+7.7$
Good cand./scene $\uparrow$	51.4	56.8	$+10.5$

Table 3: Candidate generation quality.

\uparrow

\downarrow

: higher / lower is better;

\Delta(\%)

is the relative change over the baseline.

Metric	Baseline	Ours	$\Delta(\%)$
Scene Spearman $\uparrow$	0.407	0.584	$+43.5$
Oracle recall@1 $\uparrow$	5.1%	15.4%	$+202.0$
Oracle recall@5 $\uparrow$	12.2%	29.3%	$+140.2$
Oracle rank $\downarrow$	33.9	27.7	$-18.3$
Oracle gap $\downarrow$	0.057	0.034	$-40.4$
Selected PDMS $\uparrow$	0.937	0.951	$+1.5$

Table 4: Scene-level ranking quality.

\uparrow

\downarrow

: higher / lower is better;

\Delta(\%)

is the relative improvement over the baseline.

Figure 6: Qualitative comparison of candidate ranking. Within dense bundles of geometrically similar, high-quality candidates, RRDrive selects a near-optimal trajectory where the baseline (third column) mis-ranks and selects a low-scoring one, isolating the improvement to ranking rather than to candidate generation.

Feature set	Ego-progress $\uparrow$	True PDMS $\uparrow$
Geometry only	0.369	0.371
RiskOcc only	0.174	0.177
Geometry + RiskOcc	0.535	0.533

Table 5: Within-good-candidate ranking under scene-grouped cross-validation (mean scene-wise Spearman

\uparrow

): the risk profile adds ranking information beyond trajectory geometry.

Scenario	Metric $\uparrow$	Baseline	Ours	$\Delta(\%)$
Gen-hard	Candidate PDMS	0.540	0.703	$+30.2$
Gen-hard	Good ratio	50.1%	69.8%	$+39.3$
Rank-hard	Scene Spearman	0.072	0.497	$+590.3$
Rank-hard	Selected PDMS	0.816	0.884	$+8.3$
Confu-hard	In-good Spearman	0.265	0.674	$+154.3$
Confu-hard	Oracle recall@1	3.6%	24.9%	$+591.7$

Table 6: Hardest-tercile analysis (baseline vs. Ours). All listed metrics are higher-is-better (

\uparrow

);

\Delta(\%)

is the relative improvement over the baseline.

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

Risk-Aware Occupancy for Safety-Oriented End-to-End Autonomous Driving
Sep 18, 2026Jiaxing Chen, Hengduo Zou, Yiren Zhao +1End-to-End Autonomous Driving 3D Occupancy Prediction
School of Vehicle and Mobility, Tsinghua University · The Hong Kong University of Science and Technology (Guangzhou)
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
Structured Representation Probing for Risk Anticipation in End-to-End Autonomous Driving
Oct 27, 2026Jiawei Liang, Tianrui Lou, Le Yang +3Autonomous Driving Autonomous Driving Safety Evaluation
School of Cyber Science and Technology, Shenzhen Campus of Sun Yat-sen University, Shenzhen 518107, China · School of Mathematics, Sun Yat-sen University, Guangzhou 510275, China

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

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

Jiaxing Chen, Hengduo Zou, Yiren Zhao, Bolin Gao

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

End-to-End Autonomous Driving 3D Occupancy Prediction Autonomous Driving Benchmarks

Open paper AskPDF arXiv

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.

Abstract

End-to-End Autonomous Driving

Explore similar work

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

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

Structured Representation Probing for Risk Anticipation in End-to-End Autonomous Driving