Deferred Exposure of Future Trajectories for Verifiable Reasoning in Autonomous Driving VLMs

Hugging Face Daily Papers Papers
autonomous-driving vision-language-model reasoning reinforcement-learning trajectory chain-of-thought hallucination

Summary

Introduces AD-MCQ and DEFT-RLVR, a method for verifiable reasoning in autonomous driving VLMs that defers future trajectory exposure to post-decision verification, improving reasoning faithfulness while reducing hallucinations.

Recent Vision-Language-Action (VLA) models for autonomous driving (AD) increasingly utilize chain-of-thought (CoT) supervision to enhance the reasoning capabilities of their Vision-Language Model (VLM) components, yet existing annotation pipelines commonly expose the teacher model to the logged ground-truth (GT) future trajectory. We empirically show that this induces trajectory anchoring bias: teacher models rationalize the revealed outcome rather than infer a decision from scene evidence, producing less causally faithful CoTs and substantially more severe hallucinations, especially in causally challenging scenes. Removing the GT trajectory eliminates this shortcut, but open-ended trajectory generation entangles high-level decision-making with precise geometric synthesis and low-level dynamics. To make trajectory-level driving decisions verifiable without requiring open-ended trajectory synthesis, we introduce Autonomous-Driving Multiple-Choice Question (AD-MCQ), which casts planning as selection among explicit trajectory candidates. Taking this a step further, we propose Deferred Exposure of Future Trajectories for RLVR (DEFT-RLVR) to transform future trajectories from pre-decision anchors into post-decision verification targets. Experimental results show that DEFT-RLVR improves AD reasoning while preserving or even enhancing general visual capabilities. With VLM-only inference and controllable difficulty through candidate construction, AD-MCQ provides a flexible, scalable, and extensible foundation for future research on verifiable AD reasoning.
Original Article
View Cached Full Text

Cached at: 08/04/26, 05:37 AM

Paper page - Deferred Exposure of Future Trajectories for Verifiable Reasoning in Autonomous Driving VLMs

Source: https://huggingface.co/papers/2608.01755 Published on Aug 3

·

Submitted byhttps://huggingface.co/hzxllll

hzxon Aug 4

Abstract

RecentVision-Language-Action(VLA)modelsforautonomousdriving(AD)increasinglyutilizechain-of-thought(CoT)supervisiontoenhancethereasoningcapabilitiesoftheirVision-LanguageModel(VLM)components,yetexistingannotationpipelinescommonlyexposetheteachermodeltotheloggedground-truth(GT)futuretrajectory.Weempiricallyshowthatthisinducestrajectoryanchoringbias:teachermodelsrationalizetherevealedoutcomeratherthaninferadecisionfromsceneevidence,producinglesscausallyfaithfulCoTsandsubstantiallymoreseverehallucinations,especiallyincausallychallengingscenes.RemovingtheGTtrajectoryeliminatesthisshortcut,butopen-endedtrajectorygenerationentangleshigh-leveldecision-makingwithprecisegeometricsynthesisandlow-leveldynamics.Tomaketrajectory-leveldrivingdecisionsverifiablewithoutrequiringopen-endedtrajectorysynthesis,weintroduceAutonomous-DrivingMultiple-ChoiceQuestion(AD-MCQ),whichcastsplanningasselectionamongexplicittrajectorycandidates.Takingthisastepfurther,weproposeDeferredExposureofFutureTrajectoriesforRLVR(DEFT-RLVR)totransformfuturetrajectoriesfrompre-decisionanchorsintopost-decisionverificationtargets.ExperimentalresultsshowthatDEFT-RLVRimprovesADreasoningwhilepreservingorevenenhancinggeneralvisualcapabilities.WithVLM-onlyinferenceandcontrollabledifficultythroughcandidateconstruction,AD-MCQprovidesaflexible,scalable,andextensiblefoundationforfutureresearchonverifiableADreasoning.

View arXiv pageView PDFGitHub0Add to collection

Get this paper in your agent:

hf papers read 2608\.01755

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### hzxllll/DEFT-RLVR-model-HF Image-Text-to-Text• 9B• Updatedabout 3 hours ago

Datasets citing this paper1

#### hzxllll/AD-MCQ Viewer• Updatedabout 3 hours ago • 5.5k • 7

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2608.01755 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Revealing Interpretable Failure Modes of VLMs

arXiv cs.AI

This paper introduces Revelio, a framework that systematically discovers interpretable failure modes in Vision-Language Models (VLMs) by searching over discrete concept combinations. Applied to autonomous driving and indoor robotics, it reveals previously unreported vulnerabilities that lead to crashes or safety hazards.