Research map / Vision-language and multimodal / Robot action and navigation
Grounding Spatio-Temporal Language with Transformers
Tristan Karch, Laetitia Teodorescu, Katja Hofmann, Clément Moulin-Frier, Pierre-Yves Oudeyer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Aggregate scene and language into a single vector (SFT) instead of per-object Transformer (UT/TFT) (ablation) | Transformer · Spatio-temporal sentences | F1: 0.80 -> 0.75 | Hurt |
| Use UT/TFT Transformers instead of LSTM baselines | UT/TFT · Held-out sentences, all categories | F1 | Improved |
| Add word-aggregation (WA) to Transformer models (ablation) | UT, TFT, SFT · Non-systematic held-out sentences | F1 | No clear effect |
| Add word-aggregation to UT models (ablation) | UT · Spatio-temporal concepts | F1: 2.38 | Mixed |
| Add word-aggregation (split 3, new one-to-one relation) (ablation) | UT, SFT, TFT · Split 3 'right of' | F1: -0.14, -0.15, -0.20 | Hurt |
| Add word-aggregation (split 5, past predicate grasp) (ablation) | UT, SFT, TFT · Split 5 'was grasp' | F1: 0.71, 0.45, 0.52 | Improved |
| Add word-aggregation (split 4, past spatial relation) (ablation) | UT, TFT · Split 4 'was left of' | F1 | Hurt |
| Evaluate on new observations instead of new sentences (known sentences) | LSTM baselines · Train sentences, new observations | F1: 0.79 -> 0.82 | Improved |
| Evaluate with hard negative examples instead of random negatives | Trained models · Systematic generalization splits | F1: less than 5% | Hurt |
| Increase model size with high learning rate in WA variants (ablation) | UT-WA, TFT-WA, SFT-WA · Hyperparameter search | stability | Hurt |
Closest papers by meaning
- EmbodiedBench: Comprehensive Benchmarking Multi-modal Large Language Models for Vision-Driven Embodied Agents (ICML 2025 · similarity 0.88)
- OST-Bench: Evaluating the Capabilities of MLLMs in Online Spatio-temporal Scene Understanding (NeurIPS 2025 · similarity 0.88)
- ESCA: Contextualizing Embodied Agents via Scene-Graph Generation (NeurIPS 2025 · similarity 0.88)
- ERGeoBench: A Comprehensive Benchmark for Embodied Reasoning and Geo-localization in Multimodal Large Language Models (ICML 2026 · similarity 0.88)
- Ground-Compose-Reinforce: Grounding Language in Agentic Behaviours using Limited Data (NeurIPS 2025 · similarity 0.88)
- Coarse Correspondences Boost Spatial-Temporal Reasoning in Multimodal Language Model (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
