Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Learning to Interactively Learn and Assist
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 |
|---|---|---|---|
| Train an assistant to infer tasks from a principal's physical behavior and cooperate | MAIDRQN · Experiment 1A test tasks | Joint reward: 4.0 $\pm$ 0.1 -> 4.6 $\pm$ 0.2 | Helped |
| Train an assistant to infer tasks from a principal's physical behavior and cooperate | MAIDRQN · Experiment 1B test tasks | Joint reward: 1.2 $\pm$ 0.1 -> 3.6 $\pm$ 0.1 | Helped |
| Replace the assistant LSTM with a feed-forward network (ablation) | MAIDRQN · Experiment 1A test tasks | Joint reward: 4.6 $\pm$ 0.2 -> 4.1 $\pm$ 0.1 | Hurt |
| Replace the assistant LSTM with a feed-forward network (ablation) | MAIDRQN · Experiment 1B test tasks | Joint reward: 3.6 $\pm$ 0.1 -> 2.0 $\pm$ 0.4 | Hurt |
| Restrict observations to a 1-cell window, prompting active information seeking | MAIDRQN · Experiment 2 test tasks | Joint reward | Helped |
| Allow the assistant to request task information through physical actions | MAIDRQN · Experiment 3 test tasks | Joint reward | Mixed |
| Add an assistant trained for interactive task learning | Human principal · Experiment 4 human-user games | Joint reward: 3.8 $\pm$ 0.3 -> 4.2 $\pm$ 0.4 | Helped |
| Use a shared joint value function with per-agent advantage functions instead of independent Q-learning (ablation) | Pixel-observation training (Experiment 4) · Experiment 4 training | Failure rate: 75\% -> 64\% | Helped |
| Use a shared joint value function with per-agent advantage functions instead of independent Q-learning (ablation) | Pixel-observation training (Experiment 4) · Experiment 4 training | Mean failure time: 9.7 hours -> 5.6 hours | Hurt |
Closest papers by meaning
- Interactive Learning from Activity Description (ICML 2021 · similarity 0.90)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.88)
- Synatra: Turning Indirect Knowledge into Direct Demonstrations for Digital Agents at Scale (NeurIPS 2024 · similarity 0.87)
- Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination (ICML 2025 · similarity 0.87)
- Imitation Learning from Observation with Automatic Discount Scheduling (ICLR 2024 · similarity 0.86)
- Making AI Policies Transparent to Humans through Demonstrations (AAAI 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.