An automated computer vision system enabling NAO humanoid robots to self-calibrate their joint positions using visual feedback, eliminating the need for manual calibration procedures.
System Architecture
- Monocular camera-based pose estimation using ArUco markers
- Forward kinematics modeling of NAO joint chain
- Iterative Closest Point (ICP) for pose refinement
- Automatic joint offset calculation and compensation
- Real-time feedback loop for continuous adjustment
Calibration Capabilities
- Head pan/tilt joint calibration
- Arm reach and positioning accuracy
- Leg joint alignment for stable walking
- Full-body kinematic chain optimization
Results
- Reduced calibration time from 30 minutes to under 5 minutes
- Achieved sub-degree joint accuracy
- Enabled autonomous recalibration during operation