Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Multi-Target Embodied Question Answering
Licheng Yu, Xinlei Chen, Georgia Gkioxari, Mohit Bansal, Tamara L. Berg, Dhruv Batra
Check 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 |
|---|---|---|---|
| fine-tune the navigator with RL (ablation) | Nav+Ctrl+cVQA · MT-EQA test | overall EQA accuracy (%): 54.44 -> 61.45 | Improved |
| add the controller (ablation) | Nav(RL)+cVQA · MT-EQA test | overall EQA accuracy (%): 57.40 -> 61.45 | Improved |
| use compositional VQA instead of attention-based VQA (ablation) | MT-EQA oracle with BestView images · MT-EQA oracle | overall EQA accuracy (%): 60.50 -> 74.14 | Improved |
| use controller hidden-state features instead of raw CNN features (ablation) | MT-EQA oracle with shortest paths and best-view positions · MT-EQA oracle | overall EQA accuracy (%): 74.14 -> 82.88 | Improved |
| select target-relevant frames with the controller instead of using the final state over the full sequence (ablation) | MT-EQA oracle with shortest paths · MT-EQA oracle | overall EQA accuracy (%): 52.36 -> 69.77 | Improved |
| use predicted SELECT moments rather than oracle moments (ablation) | Controller + cVQA · MT-EQA oracle | overall EQA accuracy (%): 82.88 -> 69.77 | Hurt |
| use learned navigation instead of oracle shortest paths and best-view images (ablation) | Controller + cVQA · MT-EQA | overall EQA accuracy (%): 82.88 -> 61.45 | Hurt |
| answer using question-only or prior-only predictions instead of exploring the environment | MT-EQA test · MT-EQA test | accuracy (%): 53.74 -> 61.45 | Hurt |
| answer questions with shorter ground-truth paths rather than longer paths | Nav(RL)+Ctrl+cVQA · MT-EQA test, easy vs hard | EQA accuracy (%): 72.22% over 54.92% | Improved |
Closest papers by meaning
- Knowledge-Based Embodied Question Answering (TPAMI 2023 · similarity 0.93)
- OpenEQA: Embodied Question Answering in the Era of Foundation Models (CVPR 2024 · similarity 0.92)
- IndustryEQA: Pushing the Frontiers of Embodied Question Answering in Industrial Scenarios (NeurIPS 2025 · similarity 0.88)
- Multi-modal Situated Reasoning in 3D Scenes (NeurIPS 2024 · similarity 0.87)
- Refine and Align: Confidence Calibration Through Multi-Agent Interaction in VQA (AAAI 2026 · similarity 0.87)
- Towards Learning a Generalist Model for Embodied Navigation (CVPR 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.
