Awesome-BFM-Papers

A curated list of behavior(al) foundation model (BFM) papers, articles, tutorials, slides, and projects.

Table of Contents

• Awesome-BFM-Papers
• Table of Contents
• What is the Behavior Foundation Model?
• Paper List
  • Pre-training
    • Forward-backward Representation Learning
    • Goal-conditioned Learning
    • Intrinsic Reward-driven Learning
  • Adaptation
    • Fine-tuning Techniques
    • Towards Hierarchical Control
• Datasets
• Cite Us

What is the Behavior Foundation Model?

A behavior foundation model learns broad behavior priors from large-scale and diverse behavior data, which can then be conveniently adapted to a wide range of downstream tasks.

Paper List

An overview of the pre-training pipelines and adaptation strategies for BFMs discussed in this review. The goal-conditioned learning requires an extrinsic reward function and large-scale human data, while intrinsic reward-driven learning uses intrinsic rewards generated by self-supervised tasks. In contrast, the forward-backward representation learning learns a forward embedding network (FEN) and a backward embedding network (BEN) using reward-free transitions, which can then be combined with a specific reward function to infer a policy. For adaptation strategies, BFMs can be fine-tuned through common approaches, such as full fine-tuning (FFT) and low-rank adaptation (LoRA), as well as methods like latent space adaptation, which adjust the policy by modifying the latent task vector. Beyond fine-tuning, adaptation also involves strategies for hierarchical control, where high-level planners (\textit{e.g.}, generative models like LLMs or diffusion models) process abstract goals and generate subtasks for the BFM to execute as a low-level controller, enabling complex and long-horizon task completion.

Pre-training

Forward-Backward Representation Learning

📅Year	🗃️Archive	🔤Title	📜Paper	🖥️Code
2025	arXiv	BFM-Zero: Promptable Behavioral Foundation Model for Humanoid Control Using Unsupervised Reinforcement Learning	Paper	Code
2025	ICLR	Zero-shot Whole-body Humanoid Control via Behavioral Foundation Models	Paper	Code
2024	arXiv	Finer Behavioral Foundation Models via Auto-regressive Features and Advantage Weighting	Paper	N/A
2024	NeurIPS	Fast Imitation via Behavior Foundation Models	Paper	N/A
2021	NeurIPS	Learning One Representation to Optimize All Rewards	Paper	Code
2021	arXiv	Learning Successor States and Goal-Dependent Values: A Mathematical Viewpoint	Paper	N/A

Goal-Conditioned Learning

📅Year	🗃️Archive	🔤Title	📜Paper	🖥️Code
2026	arXiv	Scaling Behavior Foundation Model for Humanoid Robots	Paper	Code
2025	arXiv	Sonic: Supersizing Motion Tracking for Natural Humanoid Whole-Body Control	Paper	Code
2025	arXiv	Track Any Motions under Any Disturbances	Paper	Code
2025	arXiv	Agility Meets Stability: Versatile Humanoid Control with Heterogeneous Data	Paper	Code
2025	arXiv	TWIST2: Scalable, Portable, and Holistic Humanoid Data Collection System	Paper	Code
2025	CoRL	TWIST: Teleoperated Whole-Body Imitation System	Paper	Code
2025	CoRL	CLONE: Closed-Loop Whole-Body Humanoid Teleoperation for Long-Horizon Tasks	Paper	Code
2025	arXiv	Behavior Foundation Model for Humanoid Robots	Paper	Code
2025	ICRA	HOVER: Versatile Neural Whole-Body Controller for Humanoid Robots	Paper	Code
2025	CVPR	InterMimic: Towards Universal Whole-Body Control for Physics-Based Human-Object Interactions	Paper	Code
2025	arXiv	ModSkill: Physical Character Skill Modularization	Paper	Code
2024	TOG	MaskedMimic: Unified Physics-Based Character Control Through Masked Motion Inpainting	Paper	Code
2024	ICLR	H-GAP: Humanoid Control with a Generalist Planner	Paper	Code
2024	SIGGRAPH	CALM: Conditional Adversarial Latent Models for Directable Virtual Characters	Paper	Code
2023	TOG	MoConVQ: Unified Physics-Based Motion Control via Scalable Discrete Representations	Paper	N/A
2023	SIGGRAPH Asia	CASE: Learning Conditional Adversarial Skill Embeddings for Physics-Based Characters	Paper	Code
2023	ICCV	PHC: Perpetual Humanoid Control for Real-Time Simulated Avatars	Paper	Code
2021	Science Robotics	TeamPlay: From Motor Control to Team Play in Simulated Humanoid Football	Paper	N/A
2023	ICML	MTM: Masked Trajectory Models for Prediction, Representation, and Control	Paper	Code
2022	TOG	ASE: Large-scale Reusable Adversarial Skill Embeddings for Physically Simulated Characters	Paper	Code

Intrinsic Reward-Driven Learning

📅Year	🗃️Archive	🔤Title	📜Paper	🖥️Code
2021	ICML	Active Pretraining with Successor Features	Paper	Code
2021	ICML	Reinforcement Learning with Prototypical Representations	Paper	Code
2020	ICML	State Entropy Maximization with Random Encoders for Efficient Exploration	Paper	Code
2019	ICLR	Exploration by Random Network Distillation	Paper	Code
2018	ICLR	Diversity is All You Need: Learning Skills without a Reward Function	Paper	Code

Adaptation

Fine-tuning Techniques

📅Year	🗃️Archive	🔤Title	📜Paper	🖥️Code
2026	arXiv	Any2Any: Efficient Cross-Embodiment Transfer for Humanoid Whole-Body Tracking	Paper	N/A
2025	arXiv	Task Tokens: A Flexible Approach to Adapting Behavior Foundation Models	Paper	N/A
2025	arXiv	Zero-Shot Adaptation of Behavioral Foundation Models to Unseen Dynamics	Paper	N/A
2025	CoRL	Fast Adaptation With Behavioral Foundation Models	Paper	N/A

Towards Hierarchical Control

📅Year	🗃️Archive	🔤Title	📜Paper	🖥️Code
2025	arXiv	SENTINEL: A Fully End-to-End Language-Action Model for Humanoid Whole Body Control	Paper	N/A
2025	arXiv	BeyondMimic: From Motion Tracking to Versatile Humanoid Control via Guided Diffusion	Paper	Code
2025	arXiv	LeVerb: Humanoid Whole-Cody Control with Latent Vision-Language Instruction	Paper	N/A
2025	arXiv	LangWBC: Language-Directed Humanoid Whole-Body Control via End-to-end Learning	Paper	N/A
2025	CVPR	Tokenhsi: Unified Synthesis of Physical Human-Scene Interactions through Task Tokenization	Paper	Code
2024	ICLR	CloSD: Closing the Loop between Simulation and Diffusion for Multi-task Character Control	Paper	Code
2024	arXiv	UniPhys: Unified Planner and Controller with Diffusion for Flexible Physics-based Character Control	Paper	Code
2023	ICLR	Unified Human-Scene Interaction via Prompted Chain-of-Contacts	Paper	Code

Datasets

📅Year	🗃️Archive	💽Dataset	🎞️Clip	⌚Hour	📜Paper	🖥️Code
2025	Humanoids	Humanoid-X	163800	240.0	Paper	Code
2025	arXiv	PHUMA	76000	73.0	Paper	Code
2025	arXiv	Motion-X++	120462	180.9	Paper	Code
2023	NIPS	Motion-X	81084	144.2	Paper	Code
2022	ECCV	PoseScript	-	-	Paper	Code
2022	CVPR	HumanML3D	14616	28.6	Paper	Code
2021	CVPR	BABEL	13220	43.5	Paper	Code
2020	TOG	LAFAN	77.0	4.6	Paper	Code
2019	ICCV	AMASS	11265	40.0	Paper	Code
2016	arXiv	KIT-ML	3911	11.2	Paper	Code

Cite Us

If this project helped your work, please cite us by

@article{yuan2025bfm,
    title={A Survey of Behavior Foundation Model: Next-Generation Whole-Body Control System of Humanoid Robots},
    author={Yuan, Mingqi and Yu, Tao and Ge, Wenqi and Yao, Xiuyong and Li, Dapeng and Wang, Huijiang and Chen, Jiayu and Li, Bo and Zhang, Wei and Zeng, Wenjun and Chen, Hua and Jin, Xin},
    journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
    year={2025},
    publisher={IEEE}
}