NeurIPS 2026 accepted papers on Reinforcement learning
395 of the 6,080 papers accepted at NeurIPS 2026 (6.5%) are on Reinforcement learning, compared with 5.6% at NeurIPS 2025. They fall into 6 topics.
See NeurIPS 2026 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest accepted papers by meaning and how crowded the spot has become.
Topics and papers
RL from feedback · 122 papers
- Jointly Reinforcing Diversity and Quality in Language Model Generations (1 citation)
- SCOPE-RL: Stable and Quantitative Control of Policy Entropy in RL Post-Training (1 citation)
- AEM: Adaptive Entropy Modulation for Multi-Turn Agentic Reinforcement Learning
- AffectGPT-RL: Revealing Roles of Reinforcement Learning in Open-Vocabulary Emotion Recognition
- AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs
- AsyncOPD: How Stale Can On-Policy Distillation Be?
- BALTO: Balanced Token-Level Policy Optimization for Hallucination Mitigation
- Beyond SFT-to-RL: Pre-alignment via Black-box On-policy Distillation for Multimodal RL
- Bilevel Optimization of Synthetic Trajectories for Multi-Turn LLM Fine-Tuning
- Block-R1: Rethinking the Role of Block Size in Multi-domain Reinforcement Learning for Diffusion Large Language Models
- Decoupling Conversational Dynamics in Full-Duplex Spoken Models through Reinforcement Learning
- Diffusion-State Policy Optimization for Masked Diffusion Language Models
and 110 more in the interactive map.
Policy gradient and actor-critic · 100 papers
- ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Scheduling (1 citation)
- A Distribution Mapping Approach to Counterfactually Fair Reinforcement Learning
- Absolute State-wise Constrained Policy Optimization: High-Probability State-wise Constraints Satisfaction
- AETDICE: Unified Framework and Offline Optimization for Nonlinear Multi-Objective RL
- Aligning Flow Map Policies with Optimal $Q$-Guidance
- Almost Sure Convergence of Linear Temporal Difference Learning with Arbitrary Features
- Breaking the Bias Barrier in Concave Multi-Objective Reinforcement Learning
- Cost-Aware Learning
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration
- Deep Double Q-learning
- Delayed homomorphic reinforcement learning for environments with delayed feedback
- Delightful Distributed Policy Gradient
and 88 more in the interactive map.
Visual and representation RL · 83 papers
- AlphaPareto: Formulaic Alpha Discovery with LLM-Guided Multi-Objective Reinforcement Learning
- Behavioral Foundation Models for Quality Diversity
- Beyond Prediction: Steering VLM Agents with Retrospective World Modeling
- Beyond Self-Play and Scale: A Behavior Benchmark for Generalization in Autonomous Driving
- CIG: Exploration via Conditional Information Gain
- DeformGen: Dynamics-Based Topology Augmentation for Deformable Manipulation Policy Learning
- Democratizing Tool Learning with Environments Fully Simulated by a Free 8B Language Model
- Distill to Think, Foresee to Act: Cognitive-Physical Reinforcement Learning for Autonomous Driving
- Embracing Evolution: A Call for Body-Control Co-Design in Embodied Humanoid Robot
- Equivariant Reinforcement Learning for Clifford Quantum Circuit Synthesis
- ForceFlow: Learning to Feel and Act via Contact-Driven Flow Matching
- Frequency‑Aware Flow Matching for Continuous and Consistent Robotic Action Generation
and 71 more in the interactive map.
Offline RL · 43 papers
- Beyond One-Size-Fits-All: Diagnosis-Driven Online Reinforcement Learning with Offline Priors
- Diffusion Guidance Is a Controllable Policy Improvement Operator
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning
- Dynamic Treatment on Networks
- From Static Policies to Adaptive Priors in Offline Reinforcement Learning
- From Weak Data to Strong Policy: Q-Targets Enable Provable In-Context Reinforcement Learning
- Generative OOD-regularized Model-based Policy Optimization
- Horizon Adaptive Offline Policy Learning via Value Stitching
- How does RL Post-training Induce Skill Composition? A Case Study on Countdown
- Localized Dynamics-Aware Domain Adaption for Off-Dynamics Offline Reinforcement Learning
- Low-Rank Adaptation for Critic Learning in Off-Policy Reinforcement Learning
- ReflectDrive-2: Reinforcement-Learning-Aligned Self-Editing for Discrete Diffusion Driving
and 31 more in the interactive map.
MDPs and planning · 27 papers
- Augmented Lagrangian Method for Last-Iterate Convergence for Constrained MDPs
- Bellman Residual Minimization for Control: Geometry, Stationarity, and Convergence
- Fast Rates for Offline Contextual Bandits with Forward-KL Regularization under Single-Policy Concentrability
- Hitting Time Isomorphism for Multi-Stage Planning with Foundation Policies
- Learning POMDP World Models from Observations with Language-Model Priors
- Multi-Environment POMDPs with Finite-Horizon Objectives
- Non-Asymptotic Best Policy Identification Guarantees in Online Reinforcement Learning
- Offline-Online Reinforcement Learning for Linear Mixture MDPs
- On the Complexity of Discounted Robust MDPs with $L_p$ Uncertainty Sets
- On the Convergence of Success Conditioning for Policy Optimization
- Persistent-Transient Policy Evaluation for Markov Chains via Minimal Peripheral Quotients
- PO-PDDL: Learning Symbolic POMDPs from Visual Demonstrations for Robot Planning Under Uncertainty
and 15 more in the interactive map.
Imitation learning · 20 papers
- Behavior Cloning is Not All You Need: The Optimality of On-Policy Distillation for Noisy Expert Feedback
- ECHO: Terminal Agents Learn World Models for Free
- Imitation from Observations with Trajectory-Level Generative Embeddings
- Imperfect World Models are Exploitable
- Inverting the Bellman Equation: From $Q$-Values to World Models
- Learning When to Stop: Selective Imitation Learning Under Arbitrary Dynamics Shift
- SkiP: When to Skip and When to Refine for Efficient Robot Manipulation
- Split the Differences, Pool the Rest: Provably Efficient Multi-Objective Imitation
- SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators
- World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry
- A Geometric Perspective on Reward Function Updates in Inverse Reinforcement Learning
- Deployment-Time Online Imitation Learning from Corrective Demonstrations
and 8 more in the interactive map.