Explainable and Resource-Efficient Spatial Reasoning in Multimodal LLMs for Decision-Critical Applications

2026-07-29

AI Summary

This work addresses the limited explainable, fine-grained spatial reasoning capabilities of multimodal large language models (MLLMs) in decision-critical scenarios, which often lead to object hallucinations and hinder auditability. To overcome this, the authors propose ByDeWay-V2, a novel framework that—without requiring additional training—integrates open-vocabulary detection (YOLO-World-L) with monocular depth estimation to explicitly model geometric relationships between objects (e.g., left-of, contains) and generate human-readable spatial predicate prompts for injection into MLLMs. This approach establishes a lightweight, interpretable spatial reasoning mechanism capable of real-time CPU execution. Evaluated on the BLINK spatial subset, it achieves a 46% higher F1 score than LDP and significantly boosts BLIP-Base’s F1 on Visual Spatial Reasoning (VSR) from near-random performance to 0.53 using only approximately 40 tokens.

Abstract

As Multimodal Large Language Models (MLLMs) are increasingly deployed in decision-critical pipelines such as robotics, embodied AI, and safety monitoring, the opacity of their spatial judgments limits operator trust and auditability. MLLMs demonstrate strong reasoning but often struggle with fine-grained spatial understanding and object hallucination. Prior work, ByDeWay, introduced Layered-Depth-Based Prompting (LDP), a training-free framework that mitigates hallucinations by structuring prompts using monocular depth estimation. However, coarse depth layering falls short in resolving object-to-object spatial relationships within the same geometric plane, such as projective ("left of", "above") and topological ("inside", "touching") relations. We propose ByDeWay-V2, which integrates explicit spatial relational context alongside depth cues, expressed as human-readable predicates that serve as auditable evidence for downstream decision support. Using an open-vocabulary object detector (YOLO-World-L), our framework computes pairwise geometric relations between detected objects and injects them as structured spatial predicates into the MLLM prompt, bridging 3D scene depth and 2D spatial semantics without any training. We evaluate ByDeWay-V2 on the Visual Spatial Reasoning (VSR) and BLINK benchmarks across multiple MLLMs, with hallucination grounding assessed via POPE. On the BLINK spatial subset, ByDeWay-V2 achieves a 46 percent relative F1 improvement over LDP for Qwen2.5-VL, and recovers BLIP-Base's spatial reasoning on VSR from near-random performance to a competitive F1 of 0.53. Our lightest configuration operates under a strict 40-token context budget on CPU, showing the framework's suitability for resource-constrained, real-time decision-support settings.

Problem

• spatial reasoning
• multimodal LLMs
• object hallucination
• explainability
• resource efficiency

Innovation

• spatial reasoning
• multimodal LLMs
• explainable AI
• hallucination mitigation
• resource-efficient inference