In a cluttered room filled with sizable ground vehicles, a compact box—approximately the size of a thick deck of cards—displayed by Australian startup Breaker, might easily go unnoticed. However, the company asserts that this unassuming device is brimming with artificial intelligence algorithms, designed for seamless integration into various unmanned platforms across a wide array of mission scenarios.
“We create robots that function like humans. Our goal is to simplify control over multiple autonomous systems across various domains,” co-founder Matthew Buffa explained to Breaking Defense during their modest booth at the biennial Land Forces conference. This sentiment is mirrored on the firm’s website, which features a video showcasing a medium-sized, four-legged robot.
Partially funded and supported by the University of New South Wales’ Founders Defence 10x accelerator program, Breaker made its debut at the Land Forces conference this year, drawing a steady stream of intrigued visitors. Buffa revealed that the company has already secured a defense customer, described only as a foreign special forces organization that has purchased several of their systems.
Buffa referred to the box as an “autonomy stack,” which incorporates six distinct AI models that can be integrated into „any autonomous platform.” He likened its functions to those of the iconic Star Wars droid R2D2, which assists Luke Skywalker in piloting, targeting, and shooting, rather than merely flooding him with information needing interpretation.
The essence of Breaker’s vision lies in fostering natural two-way communication between an operator and a drone. Buffa noted, “During times of high cognitive load and stress, humans instinctively gravitate towards voice communication.” This is particularly critical for a soldier controlling multiple drones from a ridge line. Breaker’s system allows the operator to designate one drone as the leader, enabling it to coordinate with other sensors and return for final commands.
Consider a soldier positioned along a ridge line who designates a drone as the “commander” among a swarm of ten drones. “These drones will gather sensor data across all ten platforms, and if they identify something noteworthy, such as a supply truck or a vehicle filled with armed personnel, they will notify the operator. They’ll say, ‘We’ve spotted a vehicle carrying armed individuals at this distance and direction from your current location. Should I follow them?’” explained Buffa.
Before deployment, the company trains the AIs on various mission types and threats. Once operational, Buffa stated, they “brief the drones, just like any human operator for a mission.” The drones receive clear instructions on their objectives and relevant information, provided in natural language text, including documents or scans. This information helps them make intelligent decisions.
For instance, the drone will assess whether an abandoned car near a road poses a threat. If it determines that it does not, it will refrain from unnecessarily disturbing the operator. “The operator is engaged in critical tasks. I (the drone) won’t contact him about a random car I’ve spotted, which is typical behavior for current autonomous systems,” Buffa added.
He emphasized that by offloading command responsibilities to unmanned systems, the need for a larger human presence in the field could be mitigated. This is particularly relevant considering the experiences observed on the battlefields of Ukraine, where the radio frequency signals from drone operators are susceptible to enemy counter-strikes.
“The challenges we encountered showed that for every autonomous system deployed, there is almost always at least one operator burdened with a significant amount of equipment, and communication tends to generate substantial RF noise. This has led to failures in environments like Ukraine,” Buffa noted. “If your autonomy hinges on continuous data and video, the instant that ceases, the systems become ineffective.”
