Research map / Reinforcement learning / Visual and representation RL
State-Conditioned Adversarial Subgoal Generation
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 |
|---|---|---|---|
| add state-conditioned adversarial subgoal generation to HRAC | off-policy goal-conditioned HRL · Ant Maze, Ant Maze Sparse, Ant Gather, Ant Push, Ant Fall, and Ant Maze (Image) | success rate / average episode reward | Helped |
| remove state conditioning from the discriminator (ablation) | SAGA · Ant Maze, Ant Maze Sparse, Ant Gather, Ant Push, and Ant Fall | success rate / average episode reward: 0.93$\pm$0.01 -> 0.91$\pm$0.02; 0.92$\pm$0.01 -> 0.88$\pm$0.01; 2.72$\pm$0.07 -> 2.62$\pm$0.06; 0.72$\pm$0.02 -> 0.58$\p | Hurt |
| remove state conditioning from both generator and discriminator and use a noise-input generator (ablation) | SAGA · Ant Maze, Ant Maze Sparse, Ant Gather, Ant Push, and Ant Fall | success rate / average episode reward: 0.93$\pm$0.01 -> 0.0$\pm$0.0; 0.92$\pm$0.01 -> 0.19$\pm$0.01; 2.72$\pm$0.07 -> 0.0$\pm$0.0; 0.72$\pm$0.02 -> 0.0$\pm$0.0 | Hurt |
| use HIRO rather than HRAC as the base model for SAGA (ablation) | SAGA · Ant Maze | success rate: SAGA-HIRO: 0.58$\pm$0.03; HIRO: 0.54$\pm$0.06 | Mixed |
| add adversarial subgoal generation to HIRO (ablation) | HIRO · Ant Maze Sparse, Ant Gather, Ant Push, and Ant Fall | success rate / average episode reward: Ant Maze Sparse: 0.29$\pm$0.10 -> 0.72$\pm$0.04; Ant Gather: 0.02$\pm$0.02 -> 0.18$\pm$0.15; Ant Push: 0.00$\pm$0.00 -> | Helped |
| vary the adversarial-loss coefficient among 0.0005, 0.001, and 0.0015 (ablation) | SAGA · three tasks shown in the hyperparameter analysis | performance: 0.0005; 0.001; 0.0015 | No clear effect |
Closest papers by meaning
- Robust Adversarial Reinforcement Learning (ICML 2017 · similarity 0.89)
- Generating Adjacency-Constrained Subgoals in Hierarchical Reinforcement Learning (NeurIPS 2020 · similarity 0.88)
- Learning Subgoal Representations with Slow Dynamics (ICLR 2021 · similarity 0.88)
- Hierarchical Reinforcement Learning with Timed Subgoals (NeurIPS 2021 · similarity 0.88)
- Probabilistic Subgoal Representations for Hierarchical Reinforcement Learning (ICML 2024 · similarity 0.87)
- Hierarchical Reinforcement Learning with Uncertainty-Guided Diffusional Subgoals (ICML 2025 · similarity 0.87)
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.