The Silicon Valley-based Defense Innovation Unit (DIU) operated by the US Department of Defense is seeking small unmanned surface vehicles (USVs) that are production-ready, inexpensive, and expeditionary, in addition to being capable of collaborative intercept via specialized software and/or hardware. In a separate project, the DIU is also looking for proposals that leverage deep learning to accelerate Unmanned Aerial Vehicles (UAV) swarm detection, identification, and tracking.
Launched in 2015, the DIU was one of the fruits of the Pentagon's so-called "Third Offset Strategy" that sought to leverage the private sector's innovation capacity to speed up military adoption of technological breakthroughs. The DIU's role is to identify, prototype, and scale cutting-edge, dual-use technology. In so doing, the DIU also seeks to make it easier for commercial vendors to do business with the Pentagon.
According to a solicitation posted on January 30th, the US Navy has an operational need for small USVs that can "autonomously transit hundreds of miles through contested waterspace, loitering in an assigned operating area while monitoring for maritime surface threats, and then sprinting to interdict a noncooperative, maneuvering vessel." According to the solicitation, the drone interceptors will need to operate in cohesive groups and execute complex autonomous behaviors that adapt to the dynamic, evasive movements of the pursued vessel.
The Department of Navy is seeking suppliers who can swiftly prototype and demonstrate one or more such USV interceptors, while being capable of producing them at a rate of 10 or more vehicles per month, starting from 2025. Notably, the proposed solutions will need to demonstrate a diversified and resilient manufacturing supply chain for key components such as hull, propulsion and steering, electric power, sensors, and computing systems. The small USVs in question are to be diesel-powered, and able to operate in blue-water ocean environments with an operating range 500-1000 nautical miles in moderate sea states while carrying a payload of 1,000 lbs (approximately 453.6 kg). The USVs should be able to sprint at 35 knots or faster In low sea states and loiter for several days while maintaining adequate fuel reserve for return transit.
The USVs are also expected to operate in low-visibility conditions and/or in a GNSS-denied environment, in addition to the ability to carry on the assigned missions when communications with controllers are cut off. As indicated by the solicitation, the USVs are highly desired to be RF jamming-resilient and have diversified and redundant communication that leverages high-bandwidth commercial satellite communication, 4G/5G, IP-based radios, and machine-to-machine data links and mesh networks. Other desired attributes include compatibility with various modular payloads, sensors, and effectors through compliance with the Unmanned Maritime Autonomy Architecture (UMAA) standard, the US Special Operations Command modular payload standard, and/or common commercial standards and interfaces.
To achieve what the Navy calls "Collaborative Multi-Agent Autonomy Solution", the solicitation also seeks specialized software and/or hardware that provides each sUSV with the ability to execute adaptive, cooperative behaviors and deconfliction with proximate USVs, including in crowded shipping lanes or in a GNSS-denied environment.
Using deep learning for drone swarm detection, identification, and tracking
In a separate solicitation titled Speedy UAV Swarms Detection, Identification, and Tracking using Deep Learning, the DIU indicates that the US Navy is seeking to apply deep learning on multi-sensor data fusion and exploitation system for faster UAV swarm detection, identification, and tracking via radar and infrared imagers. Those submitting applications are expected to develop and demonstrate a deep learning-based fusion methodology that can detect, identify, and track UAV swarms with a high probability of detection and a low probability of false alarm.
As pointed out by the DIU, straightforward adoption of currently fielded airspace surveillance technologies will not suffice as UAVs are much smaller in physical size and fly at lower altitudes, especially when these swarms will continuously grow in swarm size, threatening naval forces, assets, and installations. The DIU explicitly pointed to the threat posed by the Chinese military, citing reports that China is "strategizing attacking the US aircraft carrier battle groups with swarms of multi-mission UAVs during a conventional naval conflict."
Many target detection and tracking systems rely on passive medium wavelength infrared (MWIR) and long wavelength infrared (LWIR) thermal infrared cameras, and even though the performance of LWIR cameras is not easily degraded from scattering by water-based aerosols, snow, rain, fog, and clouds in the atmosphere, smaller UAVs are still hard to be detected and identified via IR imagers due to the low contrast in thermal imagery, particularly in an environment with many background noises.
As a result, the DIU notes that a single sensor solution is "impractical and ineffective" to counter drone swarms, and multiple sensors with different modalities are better suited against the threat. The DIU thus aims to develop a sensor fusion methodology based on a phased array radar system and MWIR/LWIR infrared cameras to combine data from different and orthogonal modalities, achieving the objective of detecting, identifying, and tracking drone swarm in a few seconds instead of minutes. Specifically, the DIU seeks to bring the probability of UAV swarm detection-to-track to more than 90%, reduce the probability of false alarms to less than 10% at the detection range of up to 10 km, and achieve a classification accuracy of more than 90% across the set of UAV swarms when only trained on simulated data.
To fuse multiple sensors in a UAV swarm targeting and tracking system requires the management, interpretation, and analysis of a large set of heterogeneous input data, and according to the DIU, a deep learning-based algorithm combined with multiple sensors for counter UAV swarm application has never been developed before for low-latency target detection and tracking. To validate performance claims, applicants will collect the relevant training and testing data using swarms made up of at least ten UAVs.
According to the DIU, moreover, the multi-sensor data fusion technology developed could also be applied to track anomaly detection in other domains, such as space, especially when the tracking of debris and space objects becomes important as the number of satellites in orbit grows.
