Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Information is Power: Intrinsic Control via Information Capture
Nicholas Rhinehart, Jenny Wang, Glen Berseth, John D Co-Reyes, Danijar Hafner, Chelsea Finn, Sergey Levine
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 latent-visitation entropy minimization (IC2) instead of other tested intrinsic objectives | IC2 agent · TwoRoom, VizDoom DefendTheCenter, OneRoomCapture3D | control | Improved |
| Add a belief-entropy bonus to niche creation (IC2) (ablation) | Niche Creation objective · VizDoom DefendTheCenter | monsters killed | Improved |
| Replace IC2 with Niche Creation plus Infogain (NCI) (ablation) | IC2 · OneRoomCapture3D | captured | Mixed |
| Replace IC2 with latent-state Infogain (ablation) | IC2 agent · TwoRoom, VizDoom DefendTheCenter, OneRoomCapture3D | control | Hurt |
| Replace latent-state belief objectives with observational surprise minimization (SMiRL) | Partially observed visual environments · TwoRoom, OneRoomCapture3D | control | Hurt |
| Use niche creation without an Infogain or belief-entropy bonus (ablation) | Niche Creation objective · VizDoom DefendTheCenter | monsters killed | Hurt |
| Replace IC2 with empowerment estimated from images | Partially observed visual environments · TwoRoom, OneRoomCapture3D | control | No clear effect |
| Use latent-state Infogain rather than niche creation for the intrinsic reward (ablation) | Belief-based intrinsic objective · TwoRoom, VizDoom DefendTheCenter, OneRoomCapture3D | visibility | Improved |
| Trap the box near a wall while allowing it to keep moving, rather than freeze it (ablation) | NCI agent · OneRoomCapture3D | captured | Mixed |
Closest papers by meaning
- Control What You Can: Intrinsically Motivated Task-Planning Agent (NeurIPS 2019 · similarity 0.89)
- Learning to Play With Intrinsically-Motivated, Self-Aware Agents (NeurIPS 2018 · similarity 0.88)
- Emergence of Physical Intelligence via Controllable Information Production (NeurIPS 2026 · similarity 0.88)
- Wonder Wins Ways: Curiosity-Driven Exploration through Multi-Agent Contextual Calibration (NeurIPS 2025 · similarity 0.87)
- Effective Exploration Based on the Structural Information Principles (NeurIPS 2024 · similarity 0.87)
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · 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.
