Suchet Bargoti

Suchet Bargoti builds perception and autonomy systems that help machines inspect the physical world. His career runs from counting fruit in orchards to technical leadership at Abyss Solutions and inspection and mapping at Skydio, connecting the ability to recognize what a camera sees with the harder task of making that information useful in the field.

From orchard perception to deployed inspection

Bargoti earned his PhD in perception for agricultural robotics at the University of Sydney’s Australian Centre for Field Robotics in 2017. His agricultural research combined cameras, sensors, and mapping software to build crop inventories at the level of individual trees. That granularity could help farmers direct spraying, irrigation, and harvesting toward the parts of an orchard that needed them.

With James Underwood, he investigated how computer vision could count fruit under the visual variability of real orchards. Their apple-counting research connected image segmentation with individual fruit detection and compared accumulated image counts with post-harvest counts. Their work on deep fruit detection adapted Faster R-CNN to apples, mangoes, and almonds, using image tiling to handle scenes containing many small fruits. The work addressed a practical challenge that would recur in Bargoti’s career: turning dense, imperfect observations into information someone could act on.

Bargoti joined Abyss Solutions while completing his doctorate and was working full-time as its chief technology officer by June 2017. Infrastructure inspection, including underwater environments, extended his perception work beyond agriculture and brought responsibility for technical strategy and business development.

Over an eight-year chapter at Abyss, his work developed into the engineering of deployed inspection systems. He helped build automated machine-learning pipelines, multimodal sensing and inference, and human-in-the-loop AI. These concerns widened the problem beyond recognizing an object in an image: inspection systems also needed to process different kinds of observations, carry them through repeatable inference workflows, and incorporate human judgment. The progression from orchard research to industrial inspection gave his work both a perception foundation and an operational setting.

His subsequent move to Skydio brought those concerns into drone-and-dock systems. As Director of Inspection and Mapping in 2026, he led work on autonomous drones for capturing, inspecting, and understanding critical infrastructure. Here, the machine collecting the information could also move through the environment and respond to what it found.

Making autonomy useful at fleet scale

Bargoti’s vision of drones as infrastructure centers on docked aircraft available to launch and collect information on demand. The constraint is human attention: assigning a skilled pilot to every aircraft becomes harder as fleets expand and more incidents require a response. His approach to fleet autonomy aims to let operators specify objectives while the system takes on more of the flying, perception, and coordination.

Several distinctive engineering choices shape that approach:

• Edge and cloud autonomy: Immediate flight actions happen onboard, while the cloud can support heavier inference and longer-term planning. This division makes communication part of the autonomy problem: cloud reasoning depends on receiving useful video even when bandwidth is limited.

• Maps that improve through fleet observations: Navigation combines prior information about buildings, roads, and power lines with local perception. When aircraft observe changes such as new construction, those observations can feed map updates for the wider fleet. A map becomes shared knowledge that changes as the aircraft encounter the world.

• Visual agents with flight tools: A vision-language model can interpret a request to find a white Jeep, direct a search through a drone API, and invoke tracking tools after locating the vehicle. Bargoti’s approach gives the agent existing capabilities to coordinate, reducing the need to hand-code a separate sequence of actions for every request.

Learning from operations, understanding failures

Bargoti treats field operations as a source of improvement. Differences between intended and observed behavior can inform retraining and evaluation before updated models return to the fleet. He also emphasizes removing private information from flight data and making clear to customers what they share.

His interest in learned behavior comes with a concrete engineering concern: diagnosing failures in physical systems. Fully end-to-end learning can make it difficult to understand what went wrong or guarantee dependable behavior. He supports exploring reinforcement learning and replacing hand-engineered branching rules where learning helps, while retaining the ability to investigate failures. The ambition to manage more aircraft through higher-level objectives rests on that underlying requirement: perception, planning, and flight must remain dependable as the operator’s attention moves elsewhere.