Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Interactive Visual Reasoning under Uncertainty
Manjie Xu, Guangyuan Jiang, Wei Liang, Chi Zhang, Yixin Zhu
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 |
|---|---|---|---|
| observe full episode of trials instead of context panels only | Search-Naive agent · Blicket environment | accuracy: 1.51% -> 83.80% | Improved |
| observe full episode instead of context only | Naive agent · Blicket environment | accuracy: 3.50% -> 43.62% | Improved |
| add search-based reasoning component to naive agent (ablation) | Naive agent · Blicket environment | accuracy: 43.62% -> 83.80% | Improved |
| add search-based reasoning to random trial policy (ablation) | Random agent · Blicket environment | accuracy: 1.87% -> 34.15% | Improved |
| naive trial policy instead of random trials with oracle belief (ablation) | Search-based agent · Blicket environment | accuracy: 34.15% -> 83.80% | Improved |
| recurrent instead of feed-forward policy (ablation) | DDPG · Blicket environment | accuracy: 32.47% -> 46.03% | Improved |
| recurrent instead of feed-forward policy (ablation) | TD-3 · Blicket environment | accuracy: 30.05% -> 36.83% | Improved |
| pixel input instead of symbol input (ablation) | DDPG · Blicket environment | accuracy: 32.47% -> 0.72% | Hurt |
| pixel input instead of symbol input (ablation) | TD-3 · Blicket environment | accuracy: 30.05% -> 0.31% | Hurt |
| GPT-4 instead of GPT-3.5 | LLM agent · Blicket environment | accuracy: 11% -> 26% | Improved |
| reduce to 7 objects with at most 3 blickets (ablation) | Search-Naive · Fewer objects | accuracy: 83.80% -> 100.00% | Improved |
| reduce to 7 objects with at most 3 blickets (ablation) | Naive · Fewer objects | accuracy: 43.62% -> 94.56% | Improved |
Closest papers by meaning
- I-PHYRE: Interactive Physical Reasoning (ICLR 2024 · similarity 0.86)
- Playing ZendoWorld: Challenging AI Agents on Active Visual Concept Induction (NeurIPS 2026 · similarity 0.85)
- PhysVLM-AVR: Active Visual Reasoning for Multimodal Large Language Models in Physical Environments (NeurIPS 2025 · similarity 0.85)
- Visual-O1: Understanding Ambiguous Instructions via Multi-modal Multi-turn Chain-of-thoughts Reasoning (ICLR 2025 · similarity 0.84)
- Distinguishing Performance From Competence in Evaluations of Humanlike Abstract Reasoning (NeurIPS 2026 · similarity 0.84)
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.83)
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.
