SkillBlender: Towards Versatile Humanoid Whole-Body Loco-Manipulation via Skill Blending

2025-06-11

AI Summary

Humanoid robots exhibit limited generalization in locomotion-manipulation tasks across diverse, unstructured environments; existing approaches rely on task-specific fine-tuning, compromising generality and scalability. This paper proposes SkillBlender, a hierarchical reinforcement learning framework that pretrains goal-conditioned, task-agnostic primitive skills and introduces a novel dynamic skill blending mechanism for complex task adaptation with minimal reward engineering. To mitigate reward hacking, it incorporates behavior-natural regularization, and employs optimal-control-inspired constraints to enhance motion feasibility. We further present SkillBench—the first cross-morphology, parallelized, multi-task simulation benchmark for humanoid loco-manipulation, comprising eight challenging tasks. Experiments demonstrate that SkillBlender significantly outperforms all baselines across all tasks and generalizes effectively across three distinct humanoid morphologies. The code and SkillBench benchmark will be publicly released.

Technology Category

• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Humanoid robots hold significant potential in accomplishing daily tasks across diverse environments thanks to their flexibility and human-like morphology. Recent works have made significant progress in humanoid whole-body control and loco-manipulation leveraging optimal control or reinforcement learning. However, these methods require tedious task-specific tuning for each task to achieve satisfactory behaviors, limiting their versatility and scalability to diverse tasks in daily scenarios. To that end, we introduce SkillBlender, a novel hierarchical reinforcement learning framework for versatile humanoid loco-manipulation. SkillBlender first pretrains goal-conditioned task-agnostic primitive skills, and then dynamically blends these skills to accomplish complex loco-manipulation tasks with minimal task-specific reward engineering. We also introduce SkillBench, a parallel, cross-embodiment, and diverse simulated benchmark containing three embodiments, four primitive skills, and eight challenging loco-manipulation tasks, accompanied by a set of scientific evaluation metrics balancing accuracy and feasibility. Extensive simulated experiments show that our method significantly outperforms all baselines, while naturally regularizing behaviors to avoid reward hacking, resulting in more accurate and feasible movements for diverse loco-manipulation tasks in our daily scenarios. Our code and benchmark will be open-sourced to the community to facilitate future research. Project page: https://usc-gvl.github.io/SkillBlender-web/.

Problem

• Humanoid robots need versatile loco-manipulation for diverse tasks
• Current methods require task-specific tuning, limiting scalability
• SkillBlender blends pretrained skills to minimize reward engineering

Innovation

• Hierarchical reinforcement learning for versatile humanoid control
• Dynamic blending of pretrained task-agnostic primitive skills
• SkillBench benchmark with cross-embodiment diverse tasks