As unmanned aerial vehicles (UAVs) become a staple of modern warfare, the ability to manage them at scale without requiring a dedicated operator for every single unit is essential for maximizing tactical advantages.
Palladyne AI’s SwarmOS software addresses this challenge by distributing intelligence throughout an entire swarm, utilizing principles observed in the collective behavior of ants and bees. While human operators remain central to the decision-making process, SwarmOS operates entirely at the edge on each individual drone. This eliminates the need for cloud connectivity, allowing for real-time perception, reasoning, and action even in environments where GPS is unavailable or communications are heavily contested.
In a discussion regarding the nuances of swarm capabilities and decentralized control, Palladyne AI President and CEO Ben Wolff clarified the critical technical differences between automation and autonomy.
Breaking Defense: Many organizations claim to possess swarm AI and autonomous drone capabilities, but how much of this is genuine technological advancement versus marketing? Specifically, what is the functional difference between autonomy and automation?
Wolff: From our perspective, true autonomy and swarming involve the ability for a drone to possess on-board artificial intelligence that can respond in real time to battlefield variables without constant human direction for every movement. Features like pre-programmed flight paths, waypoint coordination, or basic collision avoidance are considered automation, not autonomy. That is a fundamental distinction.
The core difference is the cognitive load: Is the human providing rigid instructions that the machine cannot deviate from, or is the drone sensing its environment and responding dynamically based on predefined parameters rather than fixed instruction sets?
Regarding the safety of armed drones, how do these concepts apply to kinetic assets?
It comes down to different levels of decision-making for specific tasks. Consider an armed drone tasked with tracking a convoy. In a traditional automated setup, a human would use a joystick to manually guide the drone. Under automation, the drone might be programmed to lock onto a specific vehicle and follow it. However, true autonomy would allow the drone to handle complex shifts in priority. For example, if a person exits the vehicle and runs, an autonomous drone could recognize the change in the situation and decide to switch targets from the car to the individual based on mission priorities.
Importantly, we are not suggesting that drones make “kill” decisions independently. Our technology handles autonomous navigation and mission prioritization. The drone might signal to the operator: “I have identified a specific target; should I pursue this for observation or execute a kinetic solution?” The final go/no-go decision for any kinetic action remains strictly with the human operator.
Palladyne AI has a long history in defense but is now pivoting toward new capabilities. Beyond autonomy, what else is your organization offering?
We spent three decades in R&D building military hardware platforms—previously operating as Sarcos Robotics and during a period as the robotics division of Raytheon. Approximately two and a half years ago, we made a strategic pivot to focus on AI. Our team has been developing decentralized AI platforms for decades, led by experts such as our head of AI, who managed AI initiatives at BAE for 15 years. We understand the battlefield’s unique challenges and have brought a novel architecture to market.
We aim to bridge the gap between human action and machine execution. There is significant confusion in the industry; simply operating without a joystick does not equate to autonomy.
How does your shift toward software, AI, and vertical integration support Department of Defense priorities?
The Pentagon is heavily focused on collaborative autonomy and increasing weapons production. Our software addresses the autonomy requirement, while our move into precision component manufacturing addresses supply chain resilience and domestic sovereignty. We now produce high-value components for critical platforms like the F-35, F-22, F-18, the Abrams tank, and various missile programs.
By manufacturing components that are traditionally difficult to source domestically, we help drive down costs, improve quality, and stabilize the supply chain for major defense primes. Furthermore, we are developing a low-cost, long-range precision effects platform designed to cost roughly one-tenth of a modern cruise missile. We achieved a transition from concept to first flight in less than six months, which is an unprecedented pace for this industry.
As a mid-tier defense prime, we combine the engineering and manufacturing rigor of a major contractor with the agility and risk tolerance of a startup.
What distinguishes SwarmOS from other systems, particularly regarding its decentralized operational capability?
Most current AI relies on the cloud because of the massive processing power required. However, small, mobile machines lack that capacity. In a communications-denied environment, even millisecond latency can be catastrophic.
Our approach is inspired by nature. First, humans can process vast amounts of sensory data while only consciously focusing on a few relevant points—like walking down stairs without thinking about every step. Second, we are inspired by the collective intelligence of social insects like ants and bees. A queen does not micromanage every member of a colony; instead, they communicate vital information—like food sources or threats—using very low-bandwidth, decentralized signals.
We have applied this “biological” philosophy to edge AI. Our algorithms are built to run on low-compute, low-cost drones without needing a central node or constant connectivity. Using onboard mesh radios, drones communicate in a hive-like manner, sharing only the essential bits of data needed to achieve the mission. A drone can essentially ask its peers, “I need more perspective on this mission; can you provide the necessary data?” This peer-to-peer collaboration is what sets us apart.
What have your recent demonstrations proven, and what does the Pentagon need to see to fully adopt this technology?
We recently completed a large-scale, real-world joint exercise. While I cannot share specific operational details, the results were significant: we successfully deployed our software across drones from multiple different manufacturers. A single soldier was able to manage the entire swarm, and the units collaborated seamlessly. The performance not only met our specifications but exceeded them.
This technology is no longer theoretical. We aren’t just presenting “toys” or PowerPoint presentations of pre-programmed flight paths. The real questions for the Pentagon are: Who makes the decisions, and when? What tasks can be offloaded to machines so humans can focus on critical objectives? Just as humans don’t consciously manage every breath, operators shouldn’t have to manage every movement of a swarm. We provide the Pentagon with a scalable, hardware-agnostic, and cost-effective solution that is ready for deployment today.
