This repository is a personal learning and reproduction project based on the paper:

"FALCON: Learning Force-Adaptive Humanoid Loco-Manipulation"
Yuanhang Zhang, Yifu Yuan, Prajwal Gurunath, et al.
[arXiv 2505.06776] | [Project Page] | [Original Code]

This repository is created for studying and reproducing the FALCON framework, which proposes a dual-agent reinforcement learning architecture for robust and generalizable humanoid loco-manipulation under external end-effector (EE) forces.

Through this project, I aim to:

• Understand the design of the dual-agent learning framework.
• Reproduce key simulation experiments.
• Analyze the torque-limit-aware force curriculum.
• Possibly extend the work to new locomotion tasks or robot platforms.