Research map / Reinforcement learning / Visual and representation RL
It Takes Four to Tango: Multiagent Self Play for Automatic Curriculum 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 |
|---|---|---|---|
| Use two symmetrized regret-based goal generators instead of one (ablation) | CuSP on most tasks · OOD random goals | success rate | Helped |
| Use two symmetrized regret-based goal generators instead of one (ablation) | CuSP · Reach OOD random goals | success rate | Mixed |
| Add entropy regularization to the regret-based goal generator (ablation) | CuSP · Most environments; OOD random-goal evaluation | success rate | Helped |
| Remove entropy regularization (ablation) | CuSP · Walker | success rate | Helped |
| Add a replay buffer for goal regrets (ablation) | CuSP goal generator · Toss | success rate | Helped |
| Add entropy regularization and a replay buffer to the regret objective (ablation) | CuSP goal generator · Toss OOD random goals | success rate | Helped |
| Add dynamic regret updates to the replay buffer (ablation) | CuSP · Hard skill-specific goals, including Walker Far Right | success rate | Helped |
| Add dynamic regret updates to the replay buffer (ablation) | CuSP · Point Mass Obstacle and Toss OOD random goals | success rate | No clear effect |
| Use two separately initialized learners to define regret rather than one learner's returns across two rollouts (ablation) | CuSP on Toss · Toss | success rate | Helped |
| Use dense rather than sparse reward for ASP+BC goal demonstration (ablation) | ASP+BC · Point Mass OOD goals | success rate | Helped |
| Train Bob with SAC rather than PPO (ablation) | ASP+BC · Control tasks | success rate | Helped |
| Train the goal generator with SAC rather than PPO or Adam (ablation) | CuSP goal generator · Toy flat regret landscape | ability to find the peak | Helped |
Closest papers by meaning
- PORTAL: Automatic Curricula Generation for Multiagent Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Heterogeneous Adversarial Play in Interactive Environments (NeurIPS 2025 · similarity 0.88)
- Learning Progress Driven Multi-Agent Curriculum (ICML 2025 · similarity 0.87)
- Diffusion-based Curriculum Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Accelerate Multi-Agent Reinforcement Learning in Zero-Sum Games with Subgame Curriculum Learning (AAAI 2024 · similarity 0.86)
- Curriculum Design for Trajectory-Constrained Agent: Compressing Chain-of-Thought Tokens in LLMs (NeurIPS 2025 · 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.