Image credit: Chenjia BaiAbstract
Executing reliable Humanoid-Object Interaction (HOI) tasks for humanoid robots is hindered by the lack of generalized control interfaces and robust closed-loop perception mechanisms. In this work, we introduce Perceptive Root-guided Humanoid-Object Interaction, Pro-HOI, a generalizable framework for robust humanoid loco-manipulation. First, we collect box-carrying motions that are suitable for real robot deployment. Based on these data, we propose a novel humanoid whole-body controller that leverages the root pose as the interface between high-level policies and low-level whole-body control, allowing seamless integration of diverse upstream perception modules. The high-level policy outputs root velocity commands and arm actions to follow end-effector position targets, while the low-level controller tracks these commands using a learned residual policy. This root-guided approach bridges the gap between high-level planning and low-level whole-body control. Real-world experiments demonstrate that our method enables the Unitree G1 humanoid robot to robustly track diverse manipulation targets in closed-loop, validating the generalization capability of our framework.