Sergey Levine

Robust Fine-tuning of Vision-Language-Action Robot Policies via Parameter Merging

Intention-Conditioned Flow Occupancy Models

Decoupled Q-Chunking

Dual Goal Representations

Learning to Reason without External Rewards

Zero-Overhead Introspection for Adaptive Test-Time Compute

Scaling Goal-conditioned Reinforcement Learning with Multistep Quasimetric Distances

Scalable Offline Model-Based RL with Action Chunks

Transitive RL: Value Learning via Divide and Conquer

Training LLM Agents to Empower Humans

Diffusion Guidance Is a Controllable Policy Improvement Operator

Natural Language Actor-Critic: Policy Iteration in Natural Language Space

Q-Learning with Adjoint Matching

Leveraging Discrete Function Decomposability for Scientific Design

RLAC: Reinforcement Learning with Adversarial Critic for Free-Form Generation Tasks

Deception in Dialogue: Evaluating and Mitigating Deceptive Behavior in Large Language Models

Dynamic Search for Inference-Time Alignment in Diffusion Models

RoboReward: A Dataset and Benchmark for Vision-Language Reward Models in Robotics

Posterior Behavioral Cloning: Pretraining BC Policies for Efficient RL Finetuning

SteerVLA: Steering Vision-Language-Action Models Toward Effective Long-Tail Driving

What really matters in matrix whitening optimizers?

Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Better

Horizon Reduction Makes RL Scalable

One Step Diffusion via Shortcut Models

Temporal Representation Alignment: Successor Features Enable Emergent Compositionality in Robot Instruction Following

Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-based Decoding

Prioritized Generative Replay

Self-Challenging Language Model Agents

Compute-Optimal Scaling for Value-Based Deep RL

Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning

OGBench: Benchmarking Offline Goal-Conditioned RL

Language Guided Skill Discovery

Reward-Guided Iterative Refinement in Diffusion Models at Test-Time with Applications to Protein and DNA Design

Real-Time Execution of Action Chunking Flow Policies

Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning

What Do Learning Dynamics Reveal About Generalization in LLM Mathematical Reasoning?

Efficient Online Reinforcement Learning Fine-Tuning Need Not Retain Offline Data

Adding Conditional Control to Diffusion Models with Reinforcement Learning

A Stable Whitening Optimizer for Efficient Neural Network Training

Reinforcement Learning with Action Chunking

Offline Goal-conditioned Reinforcement Learning with Quasimetric Representations

Scaling Test-Time Compute Without Verification or RL is Suboptimal

Flow Q-Learning

Leveraging Skills from Unlabeled Prior Data for Efficient Online Exploration

Hi Robot: Open-Ended Instruction Following with Hierarchical Vision-Language-Action Models

Fine-Tuning Discrete Diffusion Models via Reward Optimization with Applications to DNA and Protein Design

Planning without Search: Refining Frontier LLMs with Offline Goal-Conditioned RL

Leveraging Skills from Unlabeled Prior Data for Efficient Online Exploration

Proposer-Agent-Evaluator (PAE): Autonomous Skill Discovery For Foundation Model Internet Agents

Fine-Tuning of Continuous-Time Diffusion Models as Entropy-Regularized Control

Cliqueformer: Model-Based Optimization With Structured Transformers

Value-Based Deep RL Scales Predictably

LMRL Gym: Benchmarks for Multi-Turn Reinforcement Learning with Language Models

Behavioral Exploration: Learning to Explore via In-Context Adaptation

SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training

Digi-Q: Learning VLM Q-Value Functions for Training Device-Control Agents

Interactive Dialogue Agents via Reinforcement Learning with Hindsight Regenerations

Proposer-Agent-Evaluator (PAE): Autonomous Skill Discovery For Foundation Model Internet Agents

Vision-Language Models Provide Promptable Representations for Reinforcement Learning

Successor Representations Enable Emergent Compositional Instruction Following

Annotation Bootstrapping: Reinforcing Visual Pre-Training using Unlabelled Images

Unsupervised-to-Online Reinforcement Learning

ViVa: Video-Trained Value Functions for Guiding Online RL from Diverse Data

Pre-Memorization Train Accuracy Reliably Predicts Generalization in LLM Reasoning

LMRL Gym: Benchmarks for Multi-Turn Reinforcement Learning with Language Models

Defining Deception in Decision Making

Derivative-Free Guidance in Continuous and Discrete Diffusion Models with Soft Value-Based Decoding

Zero-Shot Goal Dialogue via Reinforcement Learning on Imagined Conversations