🙌 Welcome to AIR-DREAM (Decision-making Research for Empowered AI Methods) Lab code repository! AIR-DREAM is a research group at Institute for AI Industry Research (AIR), Tsinghua University. Our research focus is to develop advanced learning-based data-driven decision-making theories and practical technologies that are robust, generalizable, and deployable to tackle real-world challenges. We work on fundamental learning algorithms, embodied AI, optimization technologies for real-world AIoT systems, and data-driven decision-making tools & libraries.
Current available offline RL/IL algorithms, embodied AI models, and tools/libraries in our code repository include:
Algorithms:
- X-VLA (ICLR 2026): Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model
- xTED (AAMAS 2026): Cross-Domain Adaptation via Diffusion-Based Trajectory Editing
- Flow-Planner (NeurIPS 2025): Flow Matching-Based Autonomous Driving Planning with Advanced Interactive Behavior Modeling
- LBP (ICML 2025): Efficient Robotic Policy Learning via Latent Space Backward Planning
- UniAct (CVPR 2025): Universal Actions for Enhanced Embodied Foundation Models
- Robo-MUTUAL (ICRA 2025): Robotic Multimodal Task specifications via Unimodal Learning
- H2O+ (ICRA 2025): An Improved Framework for Hybrid Offline-and-Online RL with Dynamics Gaps
- Diffusion-Planner (ICLR 2025 oral): Diffusion-Based Planning for Autonomous Driving with Flexible Guidance
- PSEC (ICLR 2025): Skill Expansion and Composition in Parameter Space
- RSP (AAAI 2025 oral): Recursive Skip-Step Planning
- IVM (NeurIPS 2025): Instruction-Guided Visual Masking
- DecisionNCE (ICML 2024): Embodied Multimodal Representations via Implicit Preference Learning
- QPA (ICLR 2024 spotlight): Query-Policy Misalignment in Preference-Based Reinforcement Learning
- ODICE (ICLR 2024 spotlight): Revealing the Mystery of Distribution Correction Estimation via Orthogonal-gradient Update
- FISOR (ICLR 2024): Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
- PROTO: Iterative Policy Regularized Offline-to-Online Reinforcement Learning
- OMIGA (NeurIPS 2023): Offline Multi-Agent RL with Implicit Global-to-Local Value Regularization
- TSRL (NeurIPS 2023): Look Beneath the Surface: Exploiting Fundamental Symmetry for Sample-Efficient Offline RL
- SQL/EQL (ICLR 2023 oral): Offline RL with No OOD Actions: In-Sample Learning via Implicit Value Regularization
- DOGE (ICLR 2023): When Data Geometry Meets Deep Function: Generalizing Offline Reinforcement Learning
- RGM (ICLR 2023): Mind the Gap: Offline Policy Optimization for Imperfect Rewards
- H2O (NeurIPS 2022 spotlight): When to Trust Your Simulator: Dynamics-Aware Hybrid Offline-and-Online Reinforcement Learning
- POR (NeurIPS 2022 oral): A Policy-Guided Imitation Approach for Offline Reinforcement Learning
- DWBC (ICML 2022): Discriminator-Weighted Offline Imitation Learning from Suboptimal Demonstrations
- DMIL (CoRL 2022): Discriminator-Guided Model-Based Offline Imitation Learning (available from D2C)
- CPQ (AAAI 2022): Constraints Penalized Q-Learning for Safe Offline Reinforcement Learning
- DeepThermal (AAAI 2022): Combustion Optimization for Thermal Power Generating Units Using Offline Reinforcement Learning
- MOPP (IJCAI 2022): Model-Based Offline Planning with Trajectory Pruning
Tools/Libraries: