HIQL: Offline Goal-Conditioned RL with Latent States as Actions

2023/07/22 by Seohong Park, Dibya Ghosh, Park, Seohong +5 · 22 citations
Computer Science · #Artificial Intelligence (cs.AI) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Data Classification #Reinforcement Learning in Robotics #Robotics (cs.RO)

paper · pdf · doi:10.48550/arxiv.2307.11949

openalex publication_date 2023/07/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Unsupervised pre-training has recently become the bedrock for computer vision and natural language processing. In reinforcement learning (RL), goal-conditioned RL can potentially provide an analogous self-supervised approach for making use of large quantities of unlabeled (reward-free) data. However, building effective algorithms for goal-conditioned RL that can learn directly from diverse offline data is challenging, because it is hard to accurately estimate the exact value function for faraway goals. Nonetheless, goal-reaching problems exhibit structure, such that reaching distant goals entails first passing through closer subgoals. This structure can be very useful, as assessing the quality of actions for nearby goals is typically easier than for more distant goals. Based on this idea, we propose a hierarchical algorithm for goal-conditioned RL from offline data. Using one action-free value function, we learn two policies that allow us to exploit this structure: a high-level policy that treats states as actions and predicts (a latent representation of) a subgoal and a low-level policy that predicts the action for reaching this subgoal. Through analysis and didactic examples, we show how this hierarchical decomposition makes our method robust to noise in the estimated value function. We then apply our method to offline goal-reaching benchmarks, showing that our method can solve long-horizon tasks that stymie prior methods, can scale to high-dimensional image observations, and can readily make use of action-free data. Our code is available at https://seohong.me/projects/hiql/

Cited by

• Action-Sufficient Goal Representations
• Videos are Sample-Efficient Supervisions: Behavior Cloning from Videos via Latent Representations
• Planning as Descent: Goal-Conditioned Latent Trajectory Synthesis in Learned Energy Landscapes
• Goal Reaching with Eikonal-Constrained Hierarchical Quasimetric Reinforcement Learning
• Scalable Offline Model-Based RL with Action Chunks
• SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning
• Multistep Quasimetric Learning for Scalable Goal-conditioned Reinforcement Learning
• What Drives Success in Physical Planning with Joint-Embedding Predictive World Models?
• Dense and Diverse Goal Coverage in Multi Goal Reinforcement Learning
• Transitive RL: Value Learning via Divide and Conquer
• A Unified Framework for Zero-Shot Reinforcement Learning
• DAIL: Beyond Task Ambiguity for Language-Conditioned Reinforcement Learning
• Expressive Value Learning for Scalable Offline Reinforcement Learning
• Dual Goal Representations
• Test-Time Graph Search for Goal-Conditioned Reinforcement Learning
• Physics-informed Value Learner for Offline Goal-Conditioned Reinforcement Learning
• Test-time Offline Reinforcement Learning on Goal-related Experience
• Hierarchical Planning with Latent World Models
• Penalizing Infeasible Actions and Reward Scaling in Reinforcement Learning with Offline Data
• Flow-Based Single-Step Completion for Efficient and Expressive Policy Learning
• Efficient Skill Discovery via Regret-Aware Optimization
• Offline Goal-Conditioned Reinforcement Learning with Projective Quasimetric Planning

Related

• Flattening Hierarchies with Policy Bootstrapping
• Offline RL with Hierarchical Action Chunking
• Offline Goal-conditioned Reinforcement Learning with Quasimetric Representations
• Action-Sufficient Goal Representations
• OGBench: Benchmarking Offline Goal-Conditioned RL