Original article was published by Guillaume Delepine on Artificial Intelligence on Medium
Object Recognition. Have you ever wished your drone could have dodged that pesky obstacle you couldn’t see? Skydio drones can dodge obstacles that even their cameras can’t detect. Our context-aware AI algorithms can draw conclusions about what they see — for example, concluding that a cable the cameras show floating in mid-air is likely to extend in both directions. That lets them dodge obstacles that their human pilots miss, and is critical to avoiding entire categories of drone crashes.
360° Obstacle Avoidance. Have you ever lost connection to your drone and been terrified it would crash during its automated return to home? Many manual drones, like the Mavic series, for example, claim “omnidirectional obstacle avoidance” but lack full 360° sensor coverage, so without their pilots, they are prone to crash into obstacles in their blind spots. The six navigation cameras on Skydio drones have no blind spots, enabling our drones to deftly bob and weave around obstacles when flying autonomously. The Civil Air Patrol switched to Skydio 2 as their base drone, after realizing that obstacle avoidance lets them safely perform missions that would have otherwise been limited to extremely skilled pilots staying safe from obstacles at high altitude. They are able to reimagine their urban search and rescue protocols in ways that will be invaluable for disaster relief.
Motion Prediction. Have you ever gotten “stuck,” unable to fly where you want to because your collision avoidance sensors are stopping your drone in its tracks? It’s such a common problem that DJI drones even allow you to disable obstacle avoidance through Sport Mode. That’s because they don’t have motion prediction, which allows Skydio drones to plot intelligent paths around obstacles. The result is the feeling that you, the pilot, are just telling your drone where to go, while it handles the necessary twists and turns.
Advanced AI Pilot Assistance. Have you ever had trouble seeing what you wanted to see with your drone because you had to focus on flying safely? Manual drones like the DJI Phantom attempt to solve this problem by throwing people at it through two person operations — a hardly scalable approach. Our approach is to supplement the pilot with artificial intelligence, instead of another employee. I watched a bridge inspection pilot explain this to her boss once: he was skeptical that his team was using autonomy because there was still a pilot in command with a controller, but she reminded him that Skydio AI is not just letting the pilot know obstacles are present, it is actively — autonomously — intervening to navigate past them. With the confidence of an AI engine that has her back, she is able to fly closer to the assets she is inspecting, capturing angles no other drone allows her to, with greater resolution than much more expensive cameras are able to generate from farther stand-off distances.
Workflow Automation. Have you ever had to go back after a mapping flight and capture more data with a really difficult manual flight? A pilot I spoke with at a conference once told me of a time he got his Phantom stuck in the Golden Arches of a McDonald’s while trying to map the parking lot. The drone, apparently, is still up there. Mishaps are really common during autonomous mapping missions, and as a result, pilots are afraid to supplement their grid pattern flights with dangerous manual flights. We design our autonomous workflows such that the drone can assess its surroundings and make its own decisions, capturing better data more safely than a human pilot ever could. The operator is now a manager, not a pilot, which is much more scalable and safer as a drone program grows. In later blogs, we will even show you how Skydio drones can add intelligence to the scripts and make their own decisions in mid-air through Skydio 3D Scan™ and Skydio House Scan™.
This is why any truly autonomous drone requires both hardware and software to work together to provide all six of the core tenets we describe in the table above. Sure, other drones can follow a script — DJI ActiveTrack, or a photogrammetry lawnmower pattern flight. But if your hardware and software aren’t tightly coupled and built from the ground up for autonomy, these applications will always be limited, and dangerous.
So what will autonomy do for your program?
While GPS and magnetometers enabled the current generation of manual drones to evolve from toys and start to add enterprise value in the hands of skilled operators, the next era of the drone industry belongs to software-driven technology powered by breakthrough artificial intelligence that can make drones work for their pilots, instead of the other way around. In future blogs, we’ll show how this approach to autonomy translates into paradigm-shifting products that will help you take your drone program to the next level.