A new way to calibrate sensors on walking robots

The A2I-Calib++ framework helps multi-IMU systems on legged robots work more accurately.

Researchers developed A2I-Calib++, a kinematics-driven anti-noise active calibration framework. This system works for whole-body multi-IMU systems on legged robots.

The framework estimates temporal offsets, extrinsic rotations and translations, and inertial biases. It uses temporal correlation search followed by analytic spatial and bias estimation.

A2I-Calib++ has three core modules. One is a Noise-Isotropy-Inspired Trajectory that generates sufficient excitation to limit noise sensitivity. Another is a constraint-aware Adversarial Motion Priors controller. The third is a contact-constrained virtual IMU.

An inertial measurement unit (IMU) is a sensor used in multi-node systems. A rolling-contact model is used to construct the calibration reference. Kinematics refers to the robot kinematic model.

Existing calibration methods face challenges on legged robots. Routine locomotion may provide insufficient excitation. Foot slips, impacts, and vibrations introduce measurement disturbances.

Evaluations on simulation and real-world quadruped platforms show reduced noise sensitivity. The system provides improved proprioceptive odometry accuracy for legged robots.

Key facts

  • A2I-Calib++ estimates temporal offsets, extrinsic rotations, translations, and inertial biases.
  • The framework uses a contact-constrained virtual IMU to fuse calf IMUs and joint kinematics.
  • The system improves proprioceptive odometry accuracy on quadruped platforms.

All sections

Edition