Small Modular-Payload Machine Learning - Enabled USV Platforms in Support of Collaborative USV Operations
Abstract
Small low-cost USVs may be deployed from large surface ships to enable the latter to extend their reach and capabilities, particularly for intelligence, surveillance, and reconnaissance (ISR), and mine counter measure (MCM) missions in high-risk environments. Modular payload bays on such USVs allow accommodation of a range of requisite sensors as well as additional batteries for higher endurance, so that the vehicles may be reconfigured to support multiple Naval missions. Mission tasks can be distributed over a collaborative fleet of such USVs in support of more effective response to evolving adversarial threats in the maritime domain. A fleet of USVsalso enable distributed employment of a wider range of advanced sensors with potentially demanding weight and power requirements, overcoming possible small-displacement restrictions of individual USVs. Further, coordination between the USVs as well with other unmanned systems enhances the fleet performance of a task in support of the mission being pursued. The real-world situation is uncertain and continuously dynamic, and therefore difficult to model accurately. Machine learning (ML) provides an approach based on the vehicle learning from available field data in the absence or inadequacy of physical modeling, undergoing improvements over many iterations via trial-and-error interactions with the unpredictable environment in bringing the vehicle to its desired state. The trial-and-error nature of the learning process allows the USV to make autonomous decisions in adapting effectively to unforeseen circumstances, a characteristic that is vital when operating in unfamiliar surroundings, or in continually changing environmental conditions. ML is increasingly being utilized in autonomous systems with improved performance. However, acceptance of USVs in a critical Naval mission where the consequences of error can be severe requires a measure of trust that the system will perform its assigned tasks as intended. Failures can result from both hardware malfunction or from errors in the software codes and needs to be clearly identifiedas a step in developing the human operator trust, in conjunction with significant verification and validation (V&V) of the unmannedsystem. Building on an on-going ONR-funded effort under award N000141812212 that involves development of a USV for multi-domain, multi-vehicle autonomy, we propose to implement a significant level of ML-enabled autonomy on two wave-adaptive-modular-vehicle (WAM-V 16) USVs available at FAU for coordinated collaborative operations in support of selected missions of Naval interest. Previously developed ML algorithms for navigation and control, and for object detection and classification of potential obstacles or targets, would be implemented on the two USVs, in support of advancing the vehicles# autonomy and improved performance. Enabling algorithms and communication system for collaboration and coordination between the two vehicles will be developed. In support of building human trust in these USVs, a neural-network based health-monitoring system for component-level situational awareness would be developedthat would red-flag a failure of a hardware component on one USV to both USVs. Further, Explanation interfaces would be incorporated on the ML algorithms so that software failures can be identified and explained with specificity. Collaborative coordinated applications of the two USVs in conjunction with a small aerial drone will be considered in adversarial Naval scenarios, including ISR, target tracking, and swarming, with increasing levels of complexity over a three-year effort. The project will involve 1 MS student and2 undergraduates per year supervised by the two faculty mentors. The students will be engaged in STEM education and Naval engineering in support of the project.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 15, 2023
- Source ID
- N000142312426
Entities
People
- Manhar Dhanak
Organizations
- Florida Atlantic University
- Office of Naval Research
- United States Navy