Underwater Collision-Free Navigation with Domain Randomization

This lecture presents a novel mapless multi-modal end-to-end underwater 3D navigation system, leveraging domain randomization to bridge the sim-to-real gap problem. The system uses monocular depth prediction and low-cost sensors to navigate underwater environments, addressing challenges such as limited sensor configurations and image formation issues. The instructor discusses the training model's domain randomization approach, sensor configurations comparison, and real-world demonstrations in a swimming pool.