RIC-Pickup - first neonatal-lifting robotics benchmark
Infant care is one of the most physically demanding areas of healthcare, yet it remains largely untouched by automation. This gap is increasingly critical as global healthcare worker shortages and high NICU burnout rates continue to rise. To address this, we introduce RIC-Pickup—the first robotic framework focused on safely lifting an infant. The system formalizes biomechanical safety constraints for infant handling, enabling objective evaluation of robotic performance and laying the groundwork for downstream tasks such as repositioning or diaper changes. Using a teleoperation-driven data collection platform and a humanoid robot suited for NICU environments, we demonstrate through hardware trials that safe, clinically aligned infant pickup is feasible and measurable.
I led the end-to-end development of RIC-Pickup, driving all major design decisions and producing the full study and manuscript. I formalized biomechanical safety thresholds by conducting an extensive medical literature review and consulting with nurses from UCSD Medical. I then validated the ability of the Unitree G1 to maintain these criteria through hardware testing. I built the complete data-collection pipeline using OptiTrack motion capture, recording over 100 direct-human lifts and 101 teleoperated demonstrations. I then trained and evaluated an Action-Chunking Transformer (ACT) policy over joint-space sequences, achieving baseline outcomes of 100% for direct-human demonstrations, 73.3% for teleoperation, and 52.4% for the learned model, with a median model execution time of 52.1 seconds
Motivation:
The increasing demand for healthcare workers, driven by aging populations and labor shortages, presents a significant challenge for hospitals. Humanoid robots have the potential to alleviate these pressures by leveraging their human-like dexterity and adaptability to assist with medical procedures.
In this work, an exploratory study was performed into the feasibility of humanoid robots performing direct clinical tasks through teleoperation. A bi-manual teleoperation system was developed for the Unitree G1 Humanoid Robot, integrating high-fidelity pose tracking, custom grasping configurations, and an impedance controller to safely and precisely manipulate medical tools. The system is evaluated across seven diverse medical procedures, including physical examinations, emergency interventions, and precision needle tasks.
Our results demonstrate that humanoid robots can successfully replicate critical aspects of human medical assessments and interventions, with promising quantitative performance in ventilation and ultrasound-guided tasks. However, challenges remain, including limitations in force output for procedures requiring high strength and sensor sensitivity issues affecting clinical accuracy. This study highlights both the potential and current limitations of humanoid robots in hospital settings, laying the groundwork for future research in robotic healthcare integration.
My Contributions:
For this project, I designed and validated a modular tool-mount system enabling the Unitree G1 humanoid to operate a range of medical instruments during teleoperation safely. This involved developing a lightweight, rigid attachment architecture that maintained favorable center-of-mass characteristics while withstanding clinically relevant loads. I performed benchtop studies to assess structural stiffness, deflection under load, and ergonomic positioning relative to typical medical workflows. In parallel, I integrated the mounts with the robot’s bi-manual teleoperation interface, ensuring compatibility with high-precision tasks such as ultrasound guidance, airway management, and needle-based procedures. This work provided the mechanical foundation that allowed the broader system to execute diverse medical tasks with stability, accuracy, and operator confidence.