Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Ask4Help: Learning to Leverage an Expert for Embodied Tasks
Kunal Pratap Singh, Luca Weihs, Alvaro Herrasti, Jonghyun Choi, Aniruddha Kembhavi, Roozbeh Mottaghi
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 |
|---|---|---|---|
| Add a learned policy that requests expert actions when useful | EmbCLIP on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | success rate: 52.4 -> 86.3 | Improved |
| Use human experts to provide requested actions | Ask4Help on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | success rate: 55.7 -> 81.6 | Improved |
| Use a corrupted expert with epsilon 0.8 | Ask4Help on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | success rate: 52.4 -> 67.1 | Improved |
| Add learned help requests rather than use no expert | EmbCLIP on iTHOR RoomR · iTHOR RoomR validation | success rate (%): 6.90 -> 90.4 | Improved |
| Use learned requests instead of Model Confusion 0.3 | EmbCLIP on iTHOR RoomR · iTHOR RoomR validation | success rate (%): 89.9 -> 90.4 | Improved |
| Train Ask4Help with 10% rather than 25% of training scenes (ablation) | Ask4Help on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | success rate: 2% drop | Hurt |
| Train Ask4Help with 10% rather than 25% of training scenes (ablation) | Ask4Help on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | SPL: 1% | Hurt |
| Replace the pre-trained navigation agent with a random agent (ablation) | Ask4Help on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | success rate: 86.3 -> 76.44 | Hurt |
| Replace the pre-trained navigation agent with a random agent (ablation) | Ask4Help on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | expert proportion (%): 12.31 -> 98.99 | Hurt |
| Limit expert availability to 20 steps per episode | Ask4Help versus Model Confusion on RoboTHOR ObjectNav · RoboTHOR ObjectNav validation | success rate: 61% -> 70% | Improved |
Closest papers by meaning
- Retrieval-Augmented Embodied Agents (CVPR 2024 · similarity 0.85)
- Towards Learning a Generalist Model for Embodied Navigation (CVPR 2024 · similarity 0.85)
- ThinkBot: Embodied Instruction Following with Thought Chain Reasoning (ICLR 2025 · similarity 0.84)
- OpenEQA: Embodied Question Answering in the Era of Foundation Models (CVPR 2024 · similarity 0.84)
- Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models (NeurIPS 2026 · similarity 0.84)
- An Embodied Generalist Agent in 3D World (ICML 2024 · similarity 0.84)
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.
