Research map / Multi-agent systems and games / Human-AI and goal-conditioned
AvE: Assistance via Empowerment
Yuqing Du, Stas Tiomkin, Emre Kiciman, Daniel Polani, Pieter Abbeel, Anca Dragan
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 |
|---|---|---|---|
| Use empowerment instead of goal inference when the goal set is large or misspecified | Gridworld assistance · Gridworld, human in center | success rate: 50% -> 100% (LG Set); 55% -> 100% (SG Set) | Improved |
| Use empowerment rather than correctly specified goal inference | Gridworld assistance · Gridworld, human in center, SG Set | mean steps: 2.57 -> 4.8 | Hurt |
| Replace exact empowerment with the empowerment-inspired proxy (ablation) | Gridworld assistance · Gridworld, human in corner | success rate: 100% -> 99% | Hurt |
| Replace exact empowerment with the empowerment-inspired proxy (ablation) | Gridworld assistance · Gridworld, human in corner | mean steps: 5.24 -> 8.79 | Hurt |
| Add empowerment bonus to the copilot reward (ablation) | Lunar Lander copilot · Noisy pilot | successful landings: 8.5 -> 30.7 | Improved |
| Add empowerment bonus to the copilot reward (ablation) | Lunar Lander copilot · Sensor pilot | successful landings: 2.8 -> 10.8 | Improved |
| Add empowerment bonus to the copilot reward (ablation) | Lunar Lander copilot · Laggy pilot | successful landings: 38.4 -> 37.3 | No clear effect |
| Add empowerment-inspired bonus to the copilot reward (ablation) | Human-copilot Lunar Lander team · Lunar Lander user study | success rate: $17 \pm 9$ -> 33 $\pm$ 7 | Improved |
| Add empowerment-inspired bonus to the copilot reward (ablation) | Human-copilot Lunar Lander team · Lunar Lander user study | perceived task ease: 5.55 -> 6.4 | Improved |
| Increase the empowerment reward coefficient too much (ablation) | Lunar Lander copilot · Lunar Lander user study | successful landing | Hurt |
Closest papers by meaning
- Learning to Assist Humans without Inferring Rewards (NeurIPS 2024 · similarity 0.92)
- Training LLM Agents to Empower Humans (ICML 2026 · similarity 0.89)
- Shared Autonomy with IDA: Interventional Diffusion Assistance (NeurIPS 2024 · similarity 0.84)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.82)
- Proactive Agent: Shifting LLM Agents from Reactive Responses to Active Assistance (ICLR 2025 · similarity 0.82)
- AssistanceZero: Scalably Solving Assistance Games (ICML 2025 · similarity 0.82)
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.
