TL;DR
Video of one of numerous helicopter test flights at daedalean.ai to check the functionality of visual odometry and semantic terrain recognition.
At Daedalean, we work towards creating self-flying cars. This would sound familiar and, eghmmm… easy for literally everyone who is into AI nowadays: yes, you constantly hear that it is just about throwing bunch of Nice Data™ into some Deep Network™ — and you’re done! Ok, I exaggerate a lot, but the optimism is still exceptionally high.
What is usually at least under-highlighted is the amount of non-trivial work needed to get this done: gather the data, manage it, develop both ML and traditional algorithms, test them, tune them, and prove correctness. Add here a hard path of certifying such systems — and you will get an impressive list of engineering, organizational and management challenges on the way from an idea to a full realization.
Since July 2018, I’m lucky to be on a team of really professional and passionate people who work on one of the trickiest applications of AI. As part of our routines, we have to collect unique data, test our solutions under real conditions, identify wins and fails, and re-iterate. Daedalean has issued a short video snippet on how our pilots and engineers work together. Enjoy!
References
Want to discuss anything? Comments are welcome via e-mail alexey@gronskiy.com, Telegram @agronskiy or any other social media.