Stanford OnlineStanford Robotics Seminar ENGR319 | Autumn 2026 | Skydio: Autonomy for a Family of Flying Robots
Stanford Online
Remote flight deck 0:13
A live remote flight deck shows one operator logging into a system that sees all the drones and docking stations in an organization. The map marks where flight is allowed and where it is blocked by airspace limits near San Francisco. From the laptop, the speaker takes off, steers with a touchpad, and lets the autonomy system handle the hard parts.
Autonomous missions and safety 2:31
The same interface can lock onto one drone, send another to a new point, and keep a third tracking a car while the first stays on task. Drones can be sent home at the end of the work, and the system pauses one aircraft when another is too close so they do not collide. The speaker then turns back to Skydio’s wider work, from early consumer drones like R1 and Skydio 2 to the company’s full stack of dock hardware, vehicle software, autonomy, and cloud tools.
Flying robots at scale 14:30
Skydio says its drones have moved from being gadgets to infrastructure. The systems now support national security customers and allies, and sales have risen quickly after the X10 Plus docking station. The speaker compares the technical challenge to autonomous cars: a tiny body has to fly fast over dense cities while handling autonomy, aerodynamics, and reliable links. The company says it serves the backbone of civil life, with flying robots used daily to save lives and keep work going across many industries.
Autonomy and tracking 19:31
The platform is built to cut the human load in public safety and inspection. One pilot can now oversee several drones, but the goal is to move toward near zero direct piloting, with humans acting as high-level orchestrators. The system uses maps from several sources, then updates them with new data from top and bottom nav cameras, dense reconstruction, and GPS fusion. A tracking model can follow cars or people, and when line of sight is lost, a human takes over.
Inspection workflow 29:00
The software lets you load assets, choose the ones you care about, and then generates the shot sheet for them. It tells you which views to capture so you get the best coverage of the parts that matter most. You then run the mission autonomously. An RTK attachment gives a reliable pose estimate, and the captured shots can be reviewed by a human for anomalies. The hard part is knowing the exact asset location, because not every utility customer has precise grid maps.
Mapping and alerts 32:00
When locations are uncertain, the system first flies a mapping mission near the assets, detects them in the camera view, and builds a dense 3D reconstruction of the scene. It then matches those detections to the guessed asset locations and extracts their geometry, including top points and surface points, so it can place photo shots where they will cover the asset well. The same approach supports security work. Drones can fly 24/7 over defined areas, watch for people, cars, or even wooden pallets, and send alerts when something appears where it should not.
Autonomy and planning 43:00
The system now handles takeoff, landing, and flight without occlusions. You only give it a point to look at, and an optimization-based planner picks the best path while keeping the vehicle in forward flight and loitering left so it keeps line of sight. Underneath, it still uses VIO from the belly nav cameras and checkerboards on the dock to estimate pose and guide the arm for landing.
Data and deployment 45:00
Skydio says its key advantage is the loop between field use, data return, retraining, and evaluation. It draws on a large fleet dataset, plus customer data that is aggregated and de-identified. The next step is natural-language control, where you speak to the drone at a high level and it carries out missions on its own. The Q&A also covers public-safety workflows, transparency dashboards, rain limits, and cloud-based coordination.
Tracking and deployment details 58:00
The drone can reach about 100 miles an hour, and the speaker says the flight time is far higher than a quadcopter, though the exact number is not in hand. If you command it too slowly, one MPC controller under the hood starts de-transitioning and the craft moves horizontally. For defense, Xtend D uses a different radio and no 5G or cloud connectivity. GPS denial is another issue, but VIO still works at high altitude and gives a driftier estimate that is usually good enough to fly out and return. For security tracking, the team zooms in, grabs a short video snippet, and checks false positives before acting.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
