Introduction
Telepresence robots are increasingly utilized in healthcare and remote rehabilitation to bridge the distance between caregivers and patients. However, a significant "trust gap" often persists; patients may find it unsettling to be touched by a mechanical arm while the human operator is only visible on a separate 2D screen. To address this, we propose an Avatar Robot System that uses Augmented Reality (AR) to overlay a 3D model of the remote operator directly onto the physical robot. The primary challenge lies in achieving perfect visual-haptic synchronization, ensuring the patient sees the caregiver's virtual arm precisely where they feel the robot's touch.
The ONIA Solver: Bridging the Visual-Haptic Gap
The core technical hurdle in this overlay is the difference in limb proportions between humans and robots. Standard Inverse Kinematics (IK) solvers, such as Jacobian-based methods or FABRIK, are often too computationally intensive for high-frequency AR or prone to mathematical "singularities" where the model glitches.
We developed the Optimal Non-Iterative Alignment (ONIA) solver to overcome these limitations. ONIA uses a closed-form geometric approach to map a 3D human model onto an ABB YuMi robot frame in real-time.
The solver operates through several key mechanisms:
- Geometric Mapping: It treats the upper arm and forearm as vectors, aligning the shoulder's swing and twist to match the robot's elbow position.
- Dynamic Scaling: To ensure virtual joints (elbow and wrist) coincide exactly with physical robot joints, ONIA calculates a "stretch factor." This prioritizes visual overlap over strict anatomical correctness.
- Wrist Reactivity: A heuristic propagates wrist rotation back into the forearm mesh, making skin deformation look more realistic during movement.
Experimental Evaluation
The system was evaluated using a high-fidelity photogrammetry scan of a physical therapist. Performance was benchmarked against traditional solvers using 12 static arm poses and an emulated therapy session involving shoulder flexion and elbow extension.
Key Results
- Computational Efficiency: ONIA is an order of magnitude faster than Jacobian and FABRIK solvers, making it ideal for the 60Hz+ refresh rates required for stable AR rendering without "swimming" effects.
- Alignment Precision: ONIA achieved a superior "Overlay Ratio," effectively hiding the mechanical robot from the patient's perspective and maintaining zero deviation at the elbow and wrist joints.
- Robustness: Unlike iterative solvers that can fail at joint limits, ONIA remained stable throughout dynamic therapy sequences, avoiding mathematical singularities.
Discussion and Future Work
While the ONIA solver provides a robust foundation for avatar-based telepresence, some limitations remain. The dynamic scaling can lead to unnatural limb stretching (up to 1.5x) in extreme poses. Future work will focus on balancing visual overlap with anatomical plausibility and integrating torso movement. Additionally, we plan to conduct subjective user studies to quantify how this AR overlay impacts patient trust and reduces the "terror" sometimes associated with robotic medical intervention.
Summary updated