Real-time Remote Tracking and Autonomous Planning for Whale Rendezvous using Robots

TL;DR

The paper tackles real-time rendezvous with surfacing sperm whales using an autonomous UAV by integrating in situ acoustic AOA data with a whale dive model through model-based RL. It develops a multi-modal sensing and estimation architecture (acoustic AOA, VHF pulses, GMM group separation, and particle-filter localization) and an RL-driven UAV control policy that accounts for limited flight time. Field experiments in Dominica demonstrate autonomous navigation to whale-belief locations, ground-truth rendezvous near surfaced whales, and VHF-based surfacing detection extending localization during silent intervals; simulations and land tests validate robustness to sensor noise and autonomous decision-making. The work advances opportunistic, data-rich marine sensing by coupling biologically informed priors with real-time, distributed sensing and autonomous control, with future prospects for distributed VHF sensing and multi-whale rendezvous.

Abstract

We introduce a system for real-time sperm whale rendezvous at sea using an autonomous uncrewed aerial vehicle. Our system employs model-based reinforcement learning that combines in situ sensor data with an empirical whale dive model to guide navigation decisions. Key challenges include (i) real-time acoustic tracking in the presence of multiple whales, (ii) distributed communication and decision-making for robot deployments, and (iii) on-board signal processing and long-range detection from fish-trackers. We evaluate our system by conducting rendezvous with sperm whales at sea in Dominica, performing hardware experiments on land, and running simulations using whale trajectories interpolated from marine biologists' surface observations.

Figures (8)

• Figure 1: System setup for fielded deployment in Dominica. (A) The catamaran tows a linear hydrophone array to collect acoustic AOA measurements and has an antenna on its mast to detect VHF signals from an on-body fish-tracker. (B) The UAV is launched manually from the catamaran. (C) The UAV control module autonomously commands the UAV to a belief location. (D) Successful rendezvous with surfaced sperm whales.
• Figure 2: System architecture. We use a hydrophone array and PAMGuard pamguard2008 to collect acoustic AOA measurements. A local network communicates sensor information to the state estimation module and a distributed key-value store handles synchronization. Our VHF sensing payload, deployed on the catamaran and a UAV, enables long range VHF signal detection and AOA estimation. We implement a model-based RL approach that integrates sensory information with a whale dive model dswp to control a UAV.
• Figure 3: VHF sensing payload. The sensing payload consists of a Raspberry Pi that interfaces with a software-defined radio and a PX4-based flight controller to collect raw VHF signals and UAV orientation for VHF AOA computation.
• Figure 4: In situ whale rendezvous. First (A, B, C) and second rendezvous attempts (D, E, F) in Dominica. The stars indicate the group that stopped vocalizing, while triangles indicate groups that continue to vocalize. (A, D) Acoustic AOA estimates from multiple groups of whales are clustered using a Gaussian Mixture Model. (B, E) UAV's autonomous navigation trajectory. (C, F) Snapshots of the rendezvous attempt taken from the UAV's FPV camera.
• Figure 5: VHF Signal Detection for a Tagged Whale. VHF signal detection time vs. data from the tag depth sensor. We find that instantaneous detection of whale surfacing events is possible. The whale dived up to 800 meters; is clipped in the bottom plot.