Research Analyzer
← Back ICRA 2026

UniUncer: Unified Dynamic�Static Uncertainty for End-To-End Driving

Yu Gao, Jijun Wang, Zongzheng Zhang, Anqing Jiang, Yiru Wang, Yuwen Heng, Shuo Wang, HAO SUN, Zhangfeng Hu, Hao Zhao

PDF

AI summary

Key figure (auto-extracted from paper)
Jointly modeling static and dynamic uncertainty in end-to-end driving planners significantly improves trajectory accuracy and safety without adding computational overhead.
End-to-end driving Uncertainty estimation Autonomous vehicles Probabilistic planning Dynamic-static fusion

Problem

Current end-to-end autonomous driving models treat perception outputs as deterministic, ignoring inherent uncertainties in both static map elements and dynamic agents, which leads to overconfident planning and reduced robustness in complex scenarios.

Approach

The authors introduce UniUncer, a lightweight framework that converts deterministic regression heads into probabilistic Laplace regressors to estimate spatial uncertainty for both static and dynamic entities, fuses these estimates into object queries, and uses an adaptive gating mechanism to modulate historical inputs based on current uncertainty levels.

Key results

  • Reduces average L2 trajectory error by 7% on the nuScenes open-loop benchmark
  • Improves overall EPDMS by 10.8% on the NavsimV2 pseudo closed-loop benchmark
  • Adds negligible computational overhead, dropping throughput by only ~0.5 FPS
  • Ablations confirm that dynamic-agent uncertainty and the uncertainty-aware gate are both necessary for robust planning

Why it matters

Provides a plug-and-play, efficiency-preserving solution to enhance the reliability and safety of end-to-end autonomous driving systems in real-world, interaction-heavy environments.

Abstract

End-to-end (E2E) driving has become a corner- stone of both industry deployment and academic research, offering a single learnable pipeline that maps multi-sensor inputs to actions while avoiding hand-engineered modules. However, the reliability of such pipelines strongly depends on how well they handle uncertainty: sensors are noisy, se- mantics can be ambiguous, and interaction with other road users is inherently stochastic. Uncertainty also appears in multiple forms: classification vs. localization, and, crucially, in both static map elements and dynamic agents. Existing E2E approaches model only static-map uncertainty, leaving planning vulnerable to overconfident and unreliable inputs. We present UniUncer, the first lightweight, unified uncertainty framework that jointly estimates and uses uncertainty for both static and dynamic scene elements inside an E2E planner. Concretely: (1) we convert deterministic heads to probabilistic Laplace regressors that output per-vertex location and scale for vectorized static and dynamic entities; (2) we introduce an uncertainty-fusion module that encodes these parameters and injects them into object/map queries to form uncertainty- aware queries; and (3) we design an uncertainty-aware gate that adaptively modulates reliance on historical inputs (ego status or temporal perception queries) based on current uncertainty levels. The design adds minimal overhead and drops through- put by only ∼0.5 FPS while remaining plug-and-play for common E2E backbones. On nuScenes (open-loop), UniUncer reduces average L2 trajectory error by 7%. On NavsimV2 (pseudo closed-loop), it improves overall EPDMS by 10.8% and notable stage two gains in challenging, interaction-heavy scenes. Ablations confirm that dynamic-agent uncertainty and the uncertainty-aware gate are both necessary. Qualitatively, UniUncer produces well-calibrated uncertainties that encourage human-like decisions when evidence is unreliable, improving robustness without sacrificing efficiency. Our project page is: https://bradgers.github.io/Uniuncer.

Index terms

Autonomous Agents AI-Enabled Robotics

Related papers