Prev-Talk | Next-Talk | All-Talks | Talks-Sorted | MOOS-DAWG'22

Talk-05: Experimental Analysis and Validation of Collision Avoidance Algorithms under High-Traffic Scenarios in Aquatic Environments

Speakers: Mingi Jeong and Alberto Quattrini Li, Department of Computer Science, Dartmouth College

Obstacle avoidance in high-traffic waters is a key to safe and reliable maritime navigation but is still an open problem for maritime autonomy. Unlike the ground domain for self-driving cars, the aquatic domain is more challenging due to (1) unstructured traffic routes; and (2) lack of explicit rules of the road (COLREGs) under multiple simultaneous encounters. Moreover, state-of-the-art methods complying with the COLREGs assume sequential single-to-single encounters or reciprocal cooperative actions, rarely expected in real-world scenarios.

To address the above challenge, we propose a multiple obstacle avoidance (MOA) method consisting of (1) near future motion attributes-based clustering; (2) a geometric framework to identify a feasible action space; and (3) multi- objective optimization for non-myopic evasive actions. In this work, we experimentally test the proposed algorithm together with the state-of-the-art methods on both simulations and real autonomous surface vehicles (ASVs) and propose a validation framework for marine field robotics.


  • USVs
  • Collision Avoidance