mlsanityData to 14 Oct 2026

Focused Skill Discovery: Learning to Control Specific State Variables while Minimizing Side Effects

arXiv 2510.04901 ↗First posted 6 Oct 2025cs.LG cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Skills are essential for unlocking higher levels of problem solving. A common approach to discovering these skills is to learn ones that reliably reach different states, thus empowering the agent to control its environment. However, existing skill discovery algorithms often overlook the natural state variables present in many reinforcement learning problems, meaning that the discovered skills lack control of specific state variables. This can significantly hamper exploration efficiency, make skills more challenging to learn with, and lead to negative side effects in downstream tasks when the goal is under-specified. We introduce a general method that enables these skill discovery algorithms to learn focused skills -- skills that target and control specific state variables. Our approach improves state space coverage by a factor of three, unlocks new learning capabilities, and automatically avoids negative side effects in downstream tasks.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 19 of its 23 references that are in the map. First posted on arXiv 6 Oct 2025.

Builds on

19 of 23 references.

References · 19

  1. 2018Spectral Normalization for Generative Adversarial Networkscited by 1,636
  2. 2018Diversity is All You Need: Learning Skills without a Reward Functioncited by 752
  3. 2019Dynamics-Aware Unsupervised Discovery of Skillscited by 313
  4. 2016Variational Intrinsic Controlcited by 267
  5. 2019Fast Task Inference with Variational Intrinsic Successor Featurescited by 133
  6. 2018Variational Option Discovery Algorithmscited by 122
  7. 2020Explore, Discover and Learn: Unsupervised Discovery of State-Covering Skillscited by 122
  8. 2021APS: Active Pretraining with Successor Featurescited by 112
  9. 2023METRA: Scalable Unsupervised RL with Metric-Aware Abstractioncited by 86
  10. 2022Lipschitz-constrained Unsupervised Skill Discoverycited by 62
  11. 2021Hierarchical Reinforcement Learning By Discovering Intrinsic Optionscited by 58
  12. 2023Controllability-Aware Unsupervised Skill Discoverycited by 49
7 more
  1. 2019Conservative Agency via Attainable Utility Preservationcited by 35
  2. 2020Avoiding Side Effects By Considering Future Taskscited by 32
  3. 2021Unsupervised Skill Discovery with Bottleneck Option Learningcited by 28
  4. 2022Causality-driven Hierarchical Structure Discovery for Reinforcement Learningcited by 24
  5. 2020Efficient Black-Box Planning Using Macro-Actions with Focused Effectscited by 9
  6. 2024Disentangled Unsupervised Skill Discovery for Efficient Hierarchical Reinforcement Learningcited by 9
  7. 2024SkiLD: Unsupervised Skill Discovery Guided by Factor Interactionscited by 9

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Unsupervised Skill Discovery through Skill Regions Differentiation
  2. 2024When Do Skills Help Reinforcement Learning? A Theoretical Analysis of Temporal Abstractions
  3. 2025Efficient Skill Discovery via Regret-Aware Optimization
  4. 2026SUSD: Structured Unsupervised Skill Discovery through State Factorization
  5. 2026Search2Skill: Skill Distillation Beyond Knowledge Boundaries Via Rubric-Based Reinforcement Learning
  6. 2025Guiding Skill Discovery with Foundation Models
  7. 2025Human-Aligned Skill Discovery: Balancing Behaviour Exploration and Alignment
  8. 2025DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning