Humanoid robots learn to move safely on pitched roofs
Learning Slope-Adaptive Whole-Body Locomotion for Humanoid Robots in Roofing Construction
Robotics
Summary
Working on sloped roofs is hard because people have to balance and move carefully. The authors created a way for humanoid robots to learn how to walk and work on roofs by watching humans and using a 3D model of the roof. Their method helps the robot keep balance and use tools without making mistakes like slipping or hitting the roof. They tested this successfully with actions like nailing, hammering, and pushing on real and simulated roofs. This approach shows promise for using robots to help in construction tasks on tricky surfaces.
What this means in practice
- •For construction robotics teams: Train humanoid robots to perform roofing tasks safely and accurately using human-inspired motions grounded in roof geometry.
- •For industrial robot programmers: Implement scene-aware motion policies that maintain work-clearance and support stability when robots operate on sloped and uneven surfaces.
Authors
Songyang Liu, Shuai Li
Abstract
Roofing requires workers to coordinate locomotion, balance, and work-related body motions on pitched surfaces, creating a challenging application for humanoid robots. Directly retargeted human demonstrations, however, may preserve motion appearance while placing the robot's feet or hands incorrectly relative to the roof. This study presents a task-semantic scene-grounded framework for learning roofer-style whole-body motions on a Unitree G1. Human demonstrations are captured using a tracking system and retargeted to the robot, while a metric roof model supplies the spatial reference unavailable from the tracking system. A trajectory-level optimization grounds inferred support contacts and annotated work relations to the roof, and execution-aware reinforcement learning encourages the resulting policy to preserve these relations under dynamic tracking errors. The framework is evaluated through a multi-motion tracking study, a roof-pitch coverage matrix, a five-way nailgun ablation, cross-task experiments on hammering and lateral pushing, and comparisons with pure reinforcement learning and zero-shot teleoperation. Our method enables the robot to satisfy support, work-clearance, and nonpenetration criteria across all evaluated seeds. Across nailgun, hammering, and pushing, it achieves work-clearance errors between 0.256 and 0.531 cm and 3/3 successful evaluations per task. Physical experiments reproduce uphill walking, nailgun, hammering, and bending motions with mean base-frame motion errors below 80 mm. These findings establish scene-grounded human motion learning as a promising basis for construction-oriented humanoid motion primitives.