Search papers, labs, and topics across Lattice.
The paper introduces Causal Scene Narration (CSN), a method to restructure Vision-Language-Action (VLA) model inputs for autonomous driving by aligning intent and constraints, grounding them quantitatively, and separating them structurally, all at inference time without GPU cost. CSN is combined with Simplex-based runtime safety supervision and Plackett-Luce DPO training. Experiments in CARLA show that CSN improves the Driving Score by over 30% on LMDrive, with causal structure accounting for a significant portion of the gain, and that semantic safety supervision improves Infraction Score.
Autonomous driving models can get a 30% performance boost simply by restructuring text inputs to highlight causal relationships between intent and environmental constraints.
Vision-Language-Action (VLA) models for autonomous driving must integrate diverse textual inputs, including navigation commands, hazard warnings, and traffic state descriptions, yet current systems often present these as disconnected fragments, forcing the model to discover on its own which environmental constraints are relevant to the current maneuver. We introduce Causal Scene Narration (CSN), which restructures VLA text inputs through intent-constraint alignment, quantitative grounding, and structured separation, at inference time with zero GPU cost. We complement CSN with Simplex-based runtime safety supervision and training-time alignment via Plackett-Luce DPO with negative log-likelihood (NLL) regularization. A multi-town closed-loop CARLA evaluation shows that CSN improves Driving Score by +31.1% on original LMDrive and +24.5% on the preference-aligned variant. A controlled ablation reveals that causal structure accounts for 39.1% of this gain, with the remainder attributable to information content alone. A perception noise ablation confirms that CSN's benefit is robust to realistic sensing errors. Semantic safety supervision improves Infraction Score, while reactive Time-To-Collision monitoring degrades performance, demonstrating that intent-aware monitoring is needed for VLA systems.
