Depth-Constrained ASV Navigation with Deep RL and Limited Sensing
Amirhossein Zhalehmehrabi, Daniele Meli, Francesco Dal Santo, Francesco Trotti, Alessandro Farinelli
AI summary
Problem
Autonomous Surface Vehicles struggle to navigate safely in dynamic shallow waters due to strict depth constraints and highly limited sensing, typically receiving only one depth measurement per timestep.
Approach
The method combines deep reinforcement learning with a localized Gaussian Process that incrementally builds a real-time depth belief map from sparse sonar readings, enhanced by a novel confidence mechanism and gradient extrapolation.
Key results
- Localized GP regression enables uncertainty-aware depth estimation from sparse sonar data
- Novel confidence mechanism mitigates predictive variance saturation in unobserved regions
- First real-time belief update integration into an RL policy for depth-constrained navigation
- Successful zero-shot sim-to-real deployment on a physical ASV without fine-tuning
Why it matters
Enables low-cost, lightweight maritime robots to operate safely in complex coastal environments without expensive multi-sensor arrays or extensive real-world retraining.
Abstract
Autonomous Surface Vehicles (ASVs) play a crucial role in maritime operations, yet their navigation in shallow-water environments remains challenging due to dynamic disturbances and depth constraints. Traditional navigation strategies struggle with limited sensor information, making safe and efficient naviga- tion difficult. In this letter, we propose a reinforcement learning (RL) framework for ASV navigation under depth constraints, where the vehicle must reach a target while avoiding unsafe ar- eas with only a single depth measurement per timestep from a downward-facing Single Beam Echosounder (SBES). To enhance environmental awareness, we integrate Gaussian Process (GP) re- gressionintotheRLframework,enablingtheagenttoprogressively estimate a bathymetric depth map from sparse sonar readings. This approach improves decision-making by providing a richer representation of the environment. Furthermore, we demonstrate effective sim-to-real transfer, ensuring that policies generalize well to real-world aquatic conditions. Experimental results validate our method’s capability to improve ASV navigation performance while maintaining safety in challenging shallow-water environments.