EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Prospective Multimodal Memory Consolidation for Long-Horizon Vision-Language Agent Planning

Jessica Fisher

Journal of Social Sciences, Humanities and Innovation · 10.09.2026 · 2 (2) · 249-262

DOI: 10.66238/jsshi.v2i2.60 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Social Sciences, Humanities and Innovation, online erschienen am 10.09.2026. Verfasst von Jessica Fisher. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The integration of visual perception and natural language processing has significantly advanced the capabilities of artificial agents, yet long-horizon planning remains a formidable challenge due to the temporal degradation of contextual information and the sheer volume of multimodal data. This paper proposes a novel framework centered on prospective multimodal memory consolidation to address these limitations in vision-language agent planning. By mimicking human cognitive processes wherein prospective memory allows for the anticipation of future states based on consolidated past experiences, the proposed architecture dynamically filters, abstracts, and restructures high-dimensional multimodal inputs. This process ensures that agents retain semantically relevant features while discarding redundant perceptual noise over extended operational timeframes. Through a systematic evaluation across diverse simulated environments demanding complex multi-step reasoning, we demonstrate that prospective memory consolidation substantially mitigates the catastrophic forgetting commonly observed in standard transformer-based memory buffers. The methodology introduces a temporal abstraction mechanism and cross-modal alignment protocols that synergize to optimize long-term planning trajectories. Comprehensive experimental results indicate significant improvements in both task success rates and computational efficiency compared to contemporary baselines. Ultimately, this research provides a robust theoretical and empirical foundation for developing autonomous systems capable of sustained, goal-directed behavior in dynamic and unstructured environments.

Zitieren

APA

Fisher, J. (2026). Prospective Multimodal Memory Consolidation for Long-Horizon Vision-Language Agent Planning. Journal of Social Sciences, Humanities and Innovation, 2(2), 249-262. https://doi.org/10.66238/jsshi.v2i2.60

Vancouver

Fisher J. Prospective Multimodal Memory Consolidation for Long-Horizon Vision-Language Agent Planning. Journal of Social Sciences, Humanities and Innovation. 2026;2(2):249-262. doi:10.66238/jsshi.v2i2.60

BibTeX
@article{Fisher2026,
  author = {Fisher, Jessica},
  title = {Prospective Multimodal Memory Consolidation for Long-Horizon Vision-Language Agent Planning},
  journal = {Journal of Social Sciences, Humanities and Innovation},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {249--262},
  doi = {10.66238/jsshi.v2i2.60}
}
Kostenlos starten →