Reinforcement learning-based framework for whale rendezvous via autonomous sensing robots.

Area of Science:

• Robotics and Marine Biology
• Artificial Intelligence and Signal Processing

Background:

• Sperm whale prolonged dive patterns challenge biological observation efforts.
• Autonomous robot navigation for wildlife encounters requires robust tracking and rendezvous strategies.

Purpose of the Study:

• To develop an algorithmic framework for maximizing autonomous robot rendezvous opportunities with sperm whales.
• To integrate multiagent reinforcement learning for robot routing and VHF signal-based bearing estimation for whale tracking.

Main Methods:

• A framework combining a reinforcement learning-based autonomy module and a synthetic aperture radar-based VHF signal sensing module was proposed.
• The system leverages noisy bearing measurements, whale vocalizations, VHF tags, and dive behaviors for navigation.
• Field experiments used an "engineered whale" (speedboat with VHF tag) and acoustic-only tracking of untagged whales.

Main Results:

• The sensing module achieved a median bearing error of 10.55° to the VHF tag.
• The autonomy module demonstrated an 81.31% rendezvous success rate with an engineered whale at 500m using three robots.
• A 68.68% rendezvous success rate was achieved with untagged sperm whales at 1000m using two robots.

Conclusions:

• The proposed algorithmic framework effectively enhances autonomous robot rendezvous capabilities with sperm whales.
• The system shows promise for improving biological observations by enabling reliable tracking and approach.
• Validation with real-world experiments and datasets confirms the algorithm's practical applicability.