Research map / Reinforcement learning / Visual and representation RL
Learning with AMIGo: Adversarially Motivated Intrinsic Goals
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 |
|---|---|---|---|
| use AMIGo's adversarial goal-generating teacher with goal-conditioned student | IMPALA on MiniGrid · KChard, KCharder, and OMhard | mean extrinsic reward: KChard: $.00 \pm .00$ -> .54 ± .45; KCharder: $.00 \pm .00$ -> .44 ± .44; OMhard: $.00 \pm .00$ -> .17 ± .34 | Helped |
| remove extrinsic reward to the teacher (ablation) | AMIGo · MiniGrid tasks | steps to learn (millions): KCmedium: 7M -> 240M; OMmedium: 8M -> 50M; OMmedhard: 320M -> 0; KChard: 140M -> 0; KCharder: 300M -> 0; OMhard: 370M -> | Hurt |
| remove environment-change bonus (ablation) | AMIGo · MiniGrid tasks | steps to learn (millions): KCmedium: 7M -> 400M; OMmedium: 8M -> 37M; OMmedhard: 320M -> 0; KChard: 140M -> 0; KCharder: 300M -> 0; OMhard: 370M -> | Hurt |
| add a novelty bonus (ablation) | AMIGo · MiniGrid tasks | steps to learn (millions): KCmedium: 7M -> 15M; OMmedium: 8M -> 20M; OMmedhard: 320M ->350M; KChard: 140M ->100M; KCharder: 300M ->370M; OMhard: 37 | Mixed |
| replace threshold teacher reward with Gaussian reward (ablation) | AMIGo · MiniGrid tasks | steps to learn (millions): KCmedium: 7M -> 320M; OMmedium: 8M -> 60M; OMmedhard: 320M -> 0; KChard: 140M -> 0; KCharder: 300M -> 0; OMhard: 370M -> | Hurt |
| replace threshold teacher reward with Linear-Exponential reward (ablation) | AMIGo · MiniGrid tasks | steps to learn (millions): KCmedium: 7M -> 0; OMmedium: 8M -> 0; OMmedhard: 320M -> 0; KChard: 140M -> 0; KCharder: 300M -> 0; OMhard: 370M -> 0 | Hurt |
Closest papers by meaning
- Unsupervised Learning of Efficient Exploration: Pre-training Adaptive Policies via Self-Imposed Goals (ICLR 2026 · similarity 0.88)
- Adversarial Intrinsic Motivation for Reinforcement Learning (NeurIPS 2021 · similarity 0.88)
- Heterogeneous Adversarial Play in Interactive Environments (NeurIPS 2025 · similarity 0.86)
- Motif: Intrinsic Motivation from Artificial Intelligence Feedback (ICLR 2024 · similarity 0.86)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 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.