Prev-Talk | Next-Talk | All-Talks | Talks-Sorted | MOOS-DAWG'24

Talk-15: Robust Obstacle Avoidance of Autonomous Surface Vehicles Using Active Learning-augmented Intent-awareness in High-traffic Waters

Mingi Jeong(1), Arihant Chadda (2), and Alberto Quattrini Li(1)

(1) Department of Computer Science, Dartmouth College

(2) IQT Labs

Understanding nearby vehicles’ intentions is essential for effective and safe avoidance of obstacles by Autonomous Surface Vehicles (ASVs). However, these intentions are unknown to the ego ASV, as marine vessels typically do not share their plans explicitly with other vehicles. This lack of knowledge, combined with the absence of clearly marked lanes as found on roads, makes navigation challenging and results in potentially risky situations, which continues to be an open problem.

To address the challenges outlined above, we introduce a novel best-in-class method named 'active learning-augmented intent-aware obstacle avoidance’. This method is designed to handle encounters with single or multiple obstacle vehicles without requiring the ego ASV to directly communicate with other vehicles regarding their intentions. By coupling intention prediction with proactive motion planning, our proposed approach adheres to the fundamental principles of maritime conventions, specifically proactive actions as denoted in the 'Rules of the Road' (COLREGs). This enables the ego-vehicle to demonstrate good seamanship and avoid risky situations, even in a stand-on status. We have demonstrated the efficacy of this approach in simulation and real-world deployment on an ASV.

Categories:

  • USVs
  • COLREGS