Learning humanoid locomotion skills for material handling and transportation in construction sites

The increasing demand for automation in construction necessitates robotic solutions that can address labor shortages, safety concerns, and efficiency challenges. Humanoid robots, with their human-like form factor, are particularly well-suited for traversing and operating in human-centric construction environments. We presented an integrated humanoid robotic framework designed for robust locomotion and payload transportation across diverse construction surfaces. Our system combines dynamic bipedal walking with effective payload-carrying mechanisms, enabling seamless transitions between surfaces such as flat ground, ramps, stairs, narrow beams, random obstacles, and irregular trenches while carrying loads of up to 9 lb. We employed a unified training approach using reinforcement learning to achieve stable locomotion and real-time adaptation of walking styles based on the surface conditions. Experimental results demonstrate that our framework enables continuous locomotion at speeds up to 0.42 m/s across complex construction environments, showcasing the potential of humanoid robots in construction logistics.