← All speakers

Bio, Work & Ideas

Suchet Bargoti

Conference affiliation: Director of Inspection and Mapping · Skydio · 2026

Suchet Bargoti’s 2026 AI Engineer conference biography identifies him as Director of Inspection and Mapping at Skydio, leading work on autonomous drone systems for capturing, inspecting, and understanding critical infrastructure. His work spans field robotics, perception, and applied autonomy, with a focus on making drones reliable tools for collecting high-quality data in complex real-world environments.

In his presentation on Skydio’s autonomy stack, Bargoti demonstrates remote operation of multiple docked drones, including flights in San Mateo and Colorado, autonomous car tracking, and a fleet return-to-dock command. He explains how higher-level objectives can let one operator manage multiple drones, with immediate autonomous actions handled onboard and heavier inference and longer-term planning supported by the cloud. He describes shared navigation maps updated through fleet observations, tracking through occlusion, and a vision-language model using drone APIs and tracking tools to find and follow a white Jeep. He also discusses flight-data retraining and evaluation, attention to privacy and customer data sharing, and the difficulty of diagnosing failures and guaranteeing physical-system reliability with fully end-to-end learning.

1 conference talk

Key ideas

Scroll to read ↓

A live fleet demonstration leads into the engineering behind drones as infrastructure: immediate control on the aircraft, heavier reasoning in the cloud, maps refreshed by flight observations, and agents that use tools without replacing the systems responsible for safe flight.

  • Fleet operation requires drones to continue executing safely while the operator attends to another task; assigning a pilot to every aircraft preserves a staffing bottleneck.
    2:26 ↗
  • Immediate actions run on the drone, while cloud inference supports heavier reasoning and longer-term planning. Useful video under limited bandwidth helps make that split work.
    11:42 ↗
  • A shared map supports global navigation, and fleet observations can refresh it when construction changes the environment.
    13:07 ↗
  • The white Jeep example connects a language request to trajectory and tracking tools. The agent chooses how to use capabilities already exposed by the drone system.
    17:01 ↗
  • End-to-end learning can reduce manually written behavior, but difficult failure diagnosis and reliability guarantees still shape where it belongs in physical systems.
    18:31 ↗

References