Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Environmental drivers of systematicity and generalization in a situated agent
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Increase negated word/object experiences from 6 to 100 | Negation agent · Held-out negation instructions | test accuracy: 0.40 -> 0.78 | Helped |
| Train with 40 rather than 6 negated words/objects | Negation agent · Held-out negation instructions | test accuracy: 0.40 -> 0.60 | Helped |
| Use a 3D room rather than the grid world | Color-shape agent · Novel color-shape combinations | test accuracy: 0.91 -> 0.97 | Helped |
| Use a 3D Unity room rather than a 2D grid world | Putting agent · Putting held-out objects | test accuracy: 0.40 -> 0.63 | Helped |
| Add egocentric scrolling to the grid-world view (ablation) | Putting agent · Putting held-out objects | test accuracy: 0.40 -> 0.60 | Helped |
| Use a fully observable egocentric view instead of a partially observable egocentric view (ablation) | Putting agent · Putting held-out objects | test accuracy: 0.60 -> 0.64 | No clear effect |
| Make the agent visible behind the carried object (ablation) | Fixed-view grid-world putting agent · Putting held-out objects | test accuracy: 0.42 -> 0.38 | Hurt |
| Replace still-image classification with an agent that moves and changes its gaze (ablation) | Color-shape task · Novel color-shape combinations | test accuracy: 0.80 -> 1.00 | Helped |
| Remove language while retaining the object-selection task (ablation) | Grid-world language-lesion agent · Color-shape generalization task | test error | No clear effect |
| Train putting on a subset of objects and test on unseen objects | Situated agent · Putting held-out objects on the bed or tray | accuracy: over 90% | Helped |
Closest papers by meaning
- Learning Multi-Object Positional Relationships via Emergent Communication (AAAI 2024 · similarity 0.87)
- An Embodied Generalist Agent in 3D World (ICML 2024 · similarity 0.87)
- Iterated Learning Improves Compositionality in Large Vision-Language Models (CVPR 2024 · similarity 0.87)
- Attention as a Hypernetwork (ICLR 2025 · similarity 0.86)
- Evaluating alignment between humans and neural network representations in image-based learning tasks (NeurIPS 2024 · similarity 0.86)
- VLM Agents Generate Their Own Memories: Distilling Experience into Embodied Programs of Thought (NeurIPS 2024 · similarity 0.86)
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.