A Cognitive-structured Multimodal Agent that externalizes visual information into an Episodic Visual Memory and selectively reactivates relevant episodes during reasoning is proposed, enabling reinforcement learning to optimize abstraction and retrieval policies.
Abstract
Recent unified multimodal models show a single architecture can jointly perform vision/language understanding and image generation/editing. However, they repeatedly feed all historical visual and textual inputs into a shared context window, limiting long-horizon multimodal dialogue due to visual token explosion and unreliable cross-turn referencing. We propose a Cognitive-structured Multimodal Agent that externalizes visual information into an Episodic Visual Memory and selectively reactivates relevant episodes during reasoning. The agent consists of a Perceptual Abstraction Engine for structured visual abstraction, a Cognitive Retrieval Engine for cross-turn memory retrieval, and a Multimodal Executive Controller for autonomous task inference and action planning. To address the lack of turn-level retrieval supervision in existing datasets, we develop a Unified Scenario Engine that programmatically generates structured multi-turn conversations with fine-grained retrieval annotations, enabling reinforcement learning to optimize abstraction and retrieval policies. We also construct a long-horizon visual-dialogue benchmark stratified by difficulty to evaluate episodic visual recall. Our 8B agent achieves 91.4% retrieval accuracy over 20-turn sessions, surpassing 32B baselines by +8.2% while nearly halving per-turn inference time (23.1s ->12.7s). We further present the Cognitive-structured Multimodal Agent Harness (CMA-Harness), a tool-augmented deployment of the same cognitive structure integrating persistent multimodal memory, web access, image generation/editing/composition tools, and OpenAI-compatible serving. Structured memory and modular decision-making offer a more scalable, efficient paradigm for long-horizon multimodal agents than monolithic parameter scaling. Code: https://github.com/caseclose/cma-harness ; Project page: https://caseclose.github.io/cma-harness/
The Agent and Incremental Learning-based Multimodal Information Retrieval (AILMIR) framework is introduced, shifting the paradigm toward a dynamic Plan-Execute-Reflect cognitive loop, and a Non-Parametric Case-Based Memory is proposed that sediments successful reasoning trajectories, enabling efficient Domain Incremental Learning without destructive gradient updates.
Yichen Fan, Zihan Yang, Haitao Qin et al.· International Journal of Mul...· 0 citations
It is observed that the model's internal signals can indicate whether a visual step will benefit reasoning before the entire visual generation is completed, and AdaViG, a training-free adaptive visual gating method for unified multimodal reasoning, is proposed.
Embodied navigation is a core task in embodied AI. It requires comprehensive scene understanding and precise spatial reasoning. Recent vision-language models (VLMs) with strong generalization capabilities and rich commonsense knowledge have shown remarkable performance when applied to embodied navigation tasks. However, these models still encounter insufficient understanding of 3D geometry and spatial semantics when applied to real-world 3D navigation. To address this, we propose CoNav, a collaborative cross-modal reasoning framework. First, we pretrain a 3D-language model with the curriculum learning schedule and prepare a pretrained vision-language navigation agent. Next, with lightweight fine-tuning on a small 2D-3D-text corpus, the vision-language navigation agent learns to combine visual evidence with knowledge from the 3D-language model. Finally, the pretrained 3D-language model communicates with the vision-language navigation agent, enabling collaborative cross-modal reasoning and resolving ambiguities during navigation. This yields more reliable and efficient image-3D fusion for embodied navigation. CoNav introduces a new collaborative framework between a 3D-language model and a vision-language navigation agent for embodied navigation. Notably, CoNav requires only a small 2D-3D-text corpus to align 3D and 2D data. CoNav achieves clear improvements on four standard embodied navigation benchmarks (R2R, CVDN, REVERIE, SOON) and two spatial reasoning benchmarks (ScanQA, SQA3D). Moreover, Under similar success rates, it also finds shorter paths than prior methods, as measured by SPL. The results demonstrate the value of collaborative 2D-3D reasoning for embodied navigation.
Haihong Hao, Mingfei Han, Changlin Li et al.· IEEE Transactions on Pattern...· 0 citations
Document question answering over scanned pages requires two coupled abilities: (i) canonicalizing complex layouts into a faithful textual structure, and (ii) selecting and reasoning over query-relevant evidence from that structure. Most existing pipelines decouple OCR from retrieval-augmented reasoning and optimize OCR for global reconstruction, which often misaligns with evidence needs and causes brittle grounding in multi-page settings. We propose UniDocVLM, an end-to-end framework that unifies OCR and visual RAG within a single vision-language model: the model first generates a structured parse of retrieved pages, then activates question-relevant evidence from the parse to support grounded reasoning and answering. To train UniDocVLM under heterogeneous supervision, we introduce a unified JR-GRPO reinforcement learning recipe with lightweight, verifiable rewards, including format, layout-aware OCR, evidence-consistency, and answer-correctness signals, and route them to the corresponding parts of the output to improve credit assignment and reduce interference. Experiments on multi-page document QA benchmarks show that UniDocVLM yields more reliable evidence grounding and improves downstream accuracy under complex layouts.
Zong-Sheng Cao, Anran Liu, Jun Xie et al.· Proceedings of the 32nd ACM...· 0 citations
Multimodal large language models excel at passive perception but struggle with complex visual cognitive tasks requiring multi-step temporal reasoning. This degradation largely stems from the inherent ambiguity of language-based reasoning, which often fails to accurately articulate continuous visual transformations. To address this, we propose ChronoVision, a multimodal framework designed to align visual logic with latent imagery. During supervised fine-tuning, a Reconstructive Visual Head predicts the latent representation of the final transformed state, while an ROI Attention Locating module focuses the model on key visual evidence via semantic span queries. In post-training, we apply reinforcement learning with an implicit process grounding mechanism, guided by a composite reward function that evaluates outcome correctness, latent process alignment, and unsupervised visual focus. Furthermore, we introduce Vbvr-VQA, a novel dataset that evaluates temporal tracking by reformulating video reasoning into a strict image-ordering task. Experiments demonstrate that ChronoVision achieves state-of-the-art performance on Vbvr-VQA with 74.8% in-domain and 71.6% out-of-domain accuracy, alongside a strong 55.0% accuracy on IntPhys2, a highly challenging cross-domain benchmark.
An evidence-driven multimodal reasoning framework that utilizes a Bloom-inspired taxonomy as a hierarchical reasoning protocol and quantitatively analyzes the trace to make evidence usage and reasoning progression explicit is proposed.
Robert Wijaya, Ngai-Man Cheung· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
