Spatio-Temporal Grounding of Large Language Models from Perception Streams

TL;DR

This paper introduces FESTS, a framework that enhances large language models with verifiable spatio-temporal reasoning capabilities using structured video data and a novel query language.

Contribution

FESTS is a new framework that injects verifiable spatio-temporal supervision into LLMs via a structured query language and training pipeline, improving reasoning accuracy.

Findings

Training on 27k tuples improves frame-level F1 from 48.5% to 87.5%.

The model matches GPT-4.1 on complex spatio-temporal reasoning.

The approach enables spatio-temporal intelligence for Video LLMs.

Abstract

Embodied-AI agents must reason about how objects move and interact in 3-D space over time, yet existing smaller frontier Large Language Models (LLMs) still mis-handle fine-grained spatial relations, metric distances, and temporal orderings. We introduce the general framework Formally Explainable Spatio-Temporal Scenes (FESTS) that injects verifiable spatio-temporal supervision into an LLM by compiling natural-language queries into Spatial Regular Expression (SpRE) -- a language combining regular expression syntax with S4u spatial logic and extended here with universal and existential quantification. The pipeline matches each SpRE against any structured video log and exports aligned (query, frames, match, explanation) tuples, enabling unlimited training data without manual labels. Training a 3-billion-parameter model on 27k such tuples boosts frame-level F1 from 48.5% to 87.5%, matching GPT-4.1 on complex spatio-temporal reasoning while remaining two orders of magnitude smaller, and, hence, enabling spatio-temporal intelligence for Video LLM.