Connecticut-based Primordial Labs has developed a voice-based interface for small UAVs. A soldier simply speaks into a radio and tells the drone what to do.
In partnership with RedCat Holdings, Primordial is incorporating an AI-driven voice control technology called “Anura” into RedCat subsidiary Teal Drones’ Teal 2 reconnaissance UAV. The combination could give users the most intuitive form of command and control known to mankind – and RCA’s famous terrier, “Nipper” – the human voice.
“The idea is to make collaborating with robots more natural for humans,” Primordial Labs’ CEO and cofounder, Lee Ritholtz, explained in a phone interview. “It’s not necessarily making the robot more human, which I think is the traditional [view]. Rather, it’s to make the mechanism of communication more human.”
Anura is software that can run alongside command and control (C2) software on a variety of devices using standard and custom application interfaces (APIs) and protocols.
For example, a soldier operating a Teal 2 ISR drone one quarter-mile behind the front line could use an ATAK device (Android Tactical Assault Kit, basically a software-enabled chest-mounted Samsung phone) or a body-worn Tomahawk Robotics KxM controller running Anura to relay voice commands to the Teal 2 using a headset.
“They’re just talking to the [drone], giving it durable, mission type orders,” Ritholtz explains. “For example, if there are routes or points that exist on the ATAK map, they could say something like, ‘Fly route Blue then go to point Echo… Fly 45 degrees for 20 meters, then look at me.”
“Instead of flying the UAV with [a joystick] they just state a command and we carry out the rest. Anura is translating all that natural language into physical action in the real world.”
“Using natural language, drone operators can be more “heads-up” than those using tablets to issue keystroke/button commands or those flying a first-person-view (FPV) UAV via tablet-linked video according to Ritholtz.
How engaged users are for different levels of complicated missions is important. Is the user face-down flying something or can they be doing something else while we’re carrying out the flying? “
In the real world, semi-autonomous robots currently in use with the military (aerial or ground-based) are typically identified by call signs much like F-35s in a flight of two or more might be identified. Individual drones, for example, could be referred to as “Eagle Six” or “Falcon 4”. Human operators would likely interact with Teal 2 drones using such callsign identifiers via Anura to control them in the battlespace.
With the battlespace in mind, Ritholtz says Primordial Labs is constantly building a broader tactical model inside Anura based on the common operating picture provided by the C2 software it runs alongside. Applying a continually updated common operating picture to Anura allows a user to refer to environmental or tactical changes that take place in real time.
On the spur of the moment for example, an operator could tell the Teal 2 to stare at a tank that has just popped up on an ATAK map, instructing it look three-eighths of a mile to the right at its three o’clock position.
“It’s very flexible. You can instruct it as if you’re a little pilot in the UAV. You could say, ‘Drop a [map] pin where you’re looking called Echo.’ You can add things to the [common operating] map that others can see using natural language. Or you could issue an instruction relative to the entire global [map] view, saying, ‘Drop a pin 50 meters from my position and call it Delta.”
Voice interaction with semi-autonomous aerial or ground-based drones fosters more effective interaction and increased trust Ritholtz maintains. He asserts that Anura has the effect of turning C2 operators into macro-managers, not micro-managers. Primordial, he says, has seen such results in action.
Anura is already running on (unspecified) government-furnished equipment including program-of-record platforms and user devices according to Ritholtz. The interface was experimented with last September during Trident Spectre, an annual government exercise coordinated by the Naval Special Warfare Command (NSWC) that focuses on multi-agency collaboration.
Assessments have been made at several U.S. Special Operations Command (SOCOM) technical experimentation (TE) events as well. Primordial’s goal has to been to gain as much operator feedback with early versions of Anura as possible Ritholtz emphasizes.
The company’s principals came out of Lockheed MartinLMT +0.7% where Ritholtz and co-founder, Adrian Pope, built Group 2 and 3 UAVs at Lockheed’s famed Skunk Works. In 2021, they began a year-plus process of building the Anura software, taking the system out into the field and gaining feedback, some from recently-hired ex-military small UAS operators, some from the events above.
“We saw a very strong demand signal, so we productionized the system and started talking to customers across SOCOM and the Army,”
Ritholtz says. Special Operators and the Army saw value in Anura’s ability to lower the cognitive burden on drone operators.
“We’ve been talking to the Program Offices, to the requirements writers. There’s also a training burden,” Ritholtz says. “Users want to be able to choose different [drones]. When we bring Anura [to various systems] the user can just think about utilizing them rather than going through the training regimen for each one… They can just get the [drone] out there, talk to it naturally, and we take care of the rest.”
The Army’s interest has thus far translated into a direct to Phase II Small Business Innovation Research (SBIR) grant awarded last August through U.S. Army Special Operations Command (USASOC). The grant will allow Primordial Labs to deliver prototype kits for multiple, fielded Group-1 small unmanned aerial systems platforms over the next 18 months.
Primordial Labs’ formal partnership is with RedCat Holdings, essentially an investment venture designed to support companies which integrate robotic hardware and software for military, government and commercial operations like Teal Drones.
RedCat’s chief technology officer, George Matus, told me that thanks to the small size of the defense-aerospace startup community, RedCat had been aware of Primordial and the work it was doing for over a year.
“They’re one of the only vendors in the space doing this kind of thing. That, combined with us wanting to build out an ecosystem of partners, made sense. It allows us to fill a capability gap that we’ve heard about from a lot of end users… It also begins to unlock some inflection points that we’ve envisioned, allowing one operator to control many drones across domains.”
The Pentagon’s much-hyped Replicator effort will likely hinge on such capability and the potential of yoking Anura to the C2 problems that operating masses of drones suggests is a logical hypothesis.
“The way that Anura is architected, it is straightforward to integrate on a variety platforms,” Matus says. “It mostly comes down to business partnerships and negotiations.”
The Army has its own standardized C2 user interface for small UAVs called RAC-2 (Robotic and Autonomous Command and Control) which could be used with Replicator-produced drones. Anura could theoretically plug into RAC-2 but Matus says a unified C2 for the Army and other services has yet to be chosen.
Given the early stages of its development, nor has an optimal way of using Anura. The interface does not rely on keywords or a specific command script which must be memorized, speeding both operational C2 and training.
Its flexibility could be a double-edged sword however. The absence of structured commands suggests that Anura could be successfully hijacked by an adversary if an enabled device (or drone) fell into their hands.
Ritholtz admits that like the rest of the AI-enabled drone development community, Primordial Labs has struggled with this question for many years. He says Anura has protections in place including role-based authentication (user and super-user distinctions), voice-based authentication and randomly generated daily pass keys.
None of these is foolproof and a clever enemy will find workarounds. Adversaries will seek to intercept voice command signals and at minimum use their own AI to drive pattern recognition and other intelligence. As with any other technology, Anura may well be compromised by insiders, turned over to strategic competitors before it ever sees real world conflict. It will have to survive analog, cyber, and electronic warfare threats.
Obviously, the software is not immune to the connectivity vulnerabilities (cyber, EW) that can potentially disrupt all military communication and C2 on the battlefield, making voice, video or machine commands difficult or impossible. Similarly, PNT (position/navigation/timing) disruptions would degrade operators’ ability to use the Anura interface and many other C2 tools.
“The surface area of threats is massive,” Ritholtz acknowledges. “That said, there are things you can do in your software architecture from using memory-safe [programming] language to fundamentally assuming the adversary is in your system.”
Millenniums of experience in human-to-human communication have also taught us that two or more individuals can interpret the same phrase in surprisingly different ways, a risk Anura must minimize. It does so Ritholtz claims by employing domain adaptation – using a large training data set to form an understanding of the jargon, commands, common mission language and instruction used in a particular tactical setting like a land battlefield where ground forces employ ISR drones.
Anura is essentially tuned to a specific domain to reduce the amount of possible misunderstanding that could arise between operator and drone. The system offers feedback and if command is not understood, the user hears a tone in their headset. A visual readback of commands is provided as well, bolstering mutual understanding.
The software is English language-only (including accents in English) and has a degree of resilience which can be aided by structured user command language. Despite Primordial Labs’ emphasis on natural language, for efficiency and security it will have to operate within the discourse of C2 in any domain.
Ritholtz says non-English language versions of Anura may be developed. Overall, there is much more to be learned and more experience should aid in reducing if not eliminating the substantial risk of miscommunication.
Anura is set for further experimentation and evaluation in the immediate future. Primordial Labs will participate in SOCOM’s TE 24-2 at Avon Park Air Force Range, Florida in early April and TE 24-3 at Fort Liberty, North Carolina in June to name a couple.
As development progresses, Primordial Labs is considering possible alternative applications, combining Anura with ground robotics and even space systems. But it has enough development work ahead with Teal’s ISV drone to occupy it for now.
While it remains an independent startup, its status as perhaps the only company advancing an AI-enabled voice interface for military C2 suggests that success could lead to its acquisition by RedCat Holdings. But as Lee Ritholtz notes, there’s likely ample opportunity for a mature Anura.
“We’re a software company that lives and dies by its relationship with OEMs. The nice thing right now is that there are so many OEMs building great platforms.”
Source: Forbes