3D-VCD: Hallucination Mitigation in 3D-LLM Embodied Agents through Visual Contrastive Decoding

TL;DR

This paper introduces 3D-VCD, a novel inference-time contrastive decoding method that reduces hallucinations in 3D embodied agents by contrasting predictions in original and perturbed 3D scene graphs.

Contribution

It is the first to apply visual contrastive decoding at inference time for hallucination mitigation in 3D embodied reasoning, without retraining.

Findings

3D-VCD improves grounded reasoning on 3D-POPE and HEAL benchmarks.

The method suppresses language priors that are insensitive to scene grounding.

It enhances reliability of embodied agents without additional training.

Abstract

Large multimodal models are increasingly used as the reasoning core of embodied agents operating in 3D environments, yet they remain prone to hallucinations that can produce unsafe and ungrounded decisions. Existing inference-time hallucination mitigation methods largely target 2D vision-language settings and do not transfer to embodied 3D reasoning, where failures arise from object presence, spatial layout, and geometric grounding rather than pixel-level inconsistencies. We introduce 3D-VCD, the first inference-time visual contrastive decoding framework for hallucination mitigation in 3D embodied agents. 3D-VCD constructs a distorted 3D scene graph by applying semantic and geometric perturbations to object-centric representations, such as category substitutions and coordinate or extent corruption. By contrasting predictions under the original and distorted 3D contexts, our method suppresses tokens that are insensitive to grounded scene evidence and are therefore likely driven by language priors. We evaluate 3D-VCD on the 3D-POPE and HEAL benchmarks and show that it consistently improves grounded reasoning without any retraining, establishing inference-time contrastive decoding over structured 3D representations as an effective and practical route to more reliable embodied intelligence.