Humanoid Robot Locomotion in Unstructured Environments: A Survey and Perspectives

Humanoid robots are expected to operate in diverse and unpredictable environments, from cluttered indoor spaces to rough outdoor terrains. Achieving stable and agile locomotion under such conditions presents substantial challenges in motion planning and control. This survey article systematically reviews recent progress in planning and control strategies for humanoid robot locomotion in unstructured environments featuring unstructured terrains and external disturbances. We focus on existing methods in three main areas: planning, control, and unified end-to-end frameworks. Furthermore, as key components of a humanoid robot locomotion framework, perception and its integration with state estimation are briefly introduced. While classical optimization-based methods based on simplified or full-order models remain foundational, recent developments increasingly leverage learning-based policies and reactive adaptation to enhance robustness and generalization. We highlight the strengths and limitations of current approaches; identify open challenges in real-time reactivity, contact handling, and sim-to-real transfer; and outline potential future directions for integrating learning and optimization in scalable deployable locomotion systems.