POSTER
Yuming Zhang; Houtan Jebelli
Most construction robots are single-task platforms with rigid workflows, poorly suited to the diverse, contact-rich demands of dynamic job sites. Humanoid robots offer a more flexible alternative. Human-scale kinematics and whole-body mobility enable operation in existing environments without site reengineering. Yet realizing this versatility requires real-time adaptability beyond rigid automation, and given construction's unpredictability, direct human oversight remains essential, making teleoperation the most practical near-term strategy.
This research developed a context-aware humanoid teleoperation framework combining a physics-constrained full-body motion retargeting policy with a robot-view mixed reality interface. A context-adaptive control policy continuously maps operator intent to feasible humanoid actions, modulating mechanical compliance and attention allocation by operational phase. The MR interface overlays task frames, tolerance volumes, and alignment guidance into the operator's field of view, with hybrid rendering that decouples peripheral scene representation from robot camera latency to preserve perceptual continuity.
A within-subject, counterbalanced study with 20 participants evaluated the full system on a raised access floor installation task. Against a fixed-compliance baseline, the system reduced task completion time by 25.1% and ZMP variance by 44.7%. Against a no-MR baseline, completion time fell by 41.3% and positioning attempts by 50%, with significant workload and usability improvements in both comparisons. The framework advances intuitive, scalable humanoid teleoperation for complex construction tasks.