Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Interesting Object, Curious Agent: Learning Task-Agnostic Exploration
Simone Parisi, Victoria Dean, Deepak Pathak, Abhinav Gupta
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 |
|---|---|---|---|
| pre-train one exploration policy across multiple environments and transfer it | C-BET · MiniGrid transfer | task success rate | Improved |
| add environment-change rarity to state novelty (ablation) | C-BET · MiniGrid transfer | task success rate: 33.64% vs 28.35% | Improved |
| remove the change-count term (ablation) | C-BET · MiniGrid transfer | average success rate: 33.64% vs 28.35% | Hurt |
| remove the state-count term (ablation) | C-BET · MiniGrid transfer | average success rate: 33.64% vs 18.15% | Hurt |
| use both the sum and product of state and change terms (ablation) | C-BET reward · MiniGrid transfer | average success rate: 33.64% vs 31.9% | Hurt |
| use an alternative state/change reward combination (ablation) | C-BET reward · MiniGrid transfer | average success rate: 33.64% vs 30.4% | Hurt |
| count states with panoramic rather than egocentric views (ablation) | state-only count · MiniGrid transfer, three pre-training setups | average success rate: 28.3% vs 24.9% | Mixed |
| square the state-only reward (ablation) | state-only count · MiniGrid transfer | success rate | Hurt |
| encode environment changes with egocentric instead of panoramic views (ablation) | C-BET · MiniGrid transfer | success rate | Hurt |
| remove random count resets (ablation) | C-BET · MiniGrid transfer | task success rate | Hurt |
| use change-only counts instead of combined state and change counts (ablation) | C-BET · MiniGrid transfer | success rate: 33.64% vs 18.15% | Hurt |
| pre-train across multiple environments rather than a single environment (ablation) | C-BET · MiniGrid transfer | generalization to unseen environments | Improved |
Closest papers by meaning
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.88)
- Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction (ICLR 2025 · similarity 0.86)
- Wonder Wins Ways: Curiosity-Driven Exploration through Multi-Agent Contextual Calibration (NeurIPS 2025 · similarity 0.86)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.86)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.85)
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.
