Conditional diffusion method generates synthetic underwater sensor data
Conditional Diffusion Generation of DVL and Inertial Signals for Underwater Navigation
Machine Learning
Summary
Underwater robots cannot use GPS for navigation, so they rely on special sensors to estimate their movement. Collecting real sensor data from the ocean to train deep learning models is expensive and limited. The authors developed a technique that can produce realistic synthetic sensor data based on certain motion conditions, reducing the need for costly real data. They showed that training navigation models with this synthetic data performs as well as training on actual recorded data.
What this means in practice
- •For auv navigation engineers: Train underwater navigation systems without expensive sea-trial data by using synthetic DVL and inertial signal data generated conditionally on motion descriptors.
- •For autonomous marine vehicle developers: Develop and test deep learning navigation algorithms with generated sensor data when access to real-world underwater recordings is limited or impractical.
Authors
Michal Levin, Itzik Klein
Abstract
Autonomous underwater vehicles (AUVs) navigate without satellite positioning and rely on Doppler velocity log (DVL) and inertial sensors for motion estimation. Deep learning methods for processing these sensors require training data covering the motion the AUV encounters, commonly obtained through costly sea trials. To overcome this limitation, we present a conditional diffusion framework that generates synthetic DVL and inertial signals to replace real recordings in training. The generator employs an Elucidated diffusion model conditioned on three per-channel motion descriptors. Three configurations are evaluated: DVL beams alone, inertial channels alone, and all channels jointly from one model. Across thirteen trajectories of the real dataset, the generated signals followed the requested conditions with Pearson correlations between 0.72 and 0.997, transferred to trajectories withheld from generator training, and did not replicate any training window. A BeamsNet velocity estimator trained on generated signals alone, with no real recording retained, matched the velocity error obtained with the full real training set. Replacing up to all real windows with generated ones caused no measurable loss. Thus, this approach reduces the dependence of learned AUV navigation on sea-trial data, enables training when recordings are scarce, and facilitates developing and evaluating deep learning methods for underwater navigation.