Skydio 2 - track learning capability / 'practice' runs mode please

[singular.droid]

Plenty of outdoor sports have a concept of 'lap' or 'track' or 'trail' in their core - mtb/downhill, alpine skiing, competition enduro racing, motocross etc. it would be beneficial to have a chance to train drone for the lap/track/trail over a set of recorded attempts so once ready, it could deliver super-accurate following of fast-moving object throughout the whole recorded track (because it then know what to expect around itself). As user I'm ready to take consequent runs (or laps) alone same track so that drone could record several attempts and my trajectories and basically take each recorded run to learn and then navigate better through the same terrain. Ideally that after 4-5 practice runs it would be able to deliver flawless tracking without random moves where such maneuvers are not possible.

Imagine I'm riding a mtb downhill trail and I can sacrifice half of a day to make a drone learn the trail: I will do first descend at 10kmh and if/when the drone lost my track I would stop and return it to the trail. Then next run I descend the same trail at 15kmh again making drone learn it at higher speed, then at 20, then at 30, then at full speed. Due to the fact drone has recorded all these attempts with my trajectories and own trajectories over several runs it could basically optimize its own route and probably record 'safe tunnel' for following and flying and as consequence, move faster and better. This mode will help a lot if an object of tracking will move at 30-40-50kmh median speed alone the track and surroundings are not easy to fly through (forest etc). This mode will also so help greatly if I host some event and record each participant attempt - I can pre-practice drone day before and then next day it will capture participants way better than 'out of blind flight following them'

 

[skydioconvert]

Interesting concept but I suspect the technical challenges of having pre-recorded flight paths available to select from may not not be feasible with the current technology.

The memory and processing power needed to record the data of multiple runs, determine the best run and then store even one such optimum path is probably well outside of current capabilities.

These chips already run really hot and to task them even further is probably a can of worms the Skydio engineers don't even want to think about opening.

 

[singular.droid]

Well, in my stupid programming eyes it is just a task for post flight optimization. It can be supervised ie you run it manually as merge of all trajectories once you get the logs out of device with help of airwolf or similar software. In this case think of it as video editor where you have several tracks running in parallel and you cut out faulty parts leaving only best ones. Or all of this could be unsupervised when you turn the drone on, select flights via mobile app and request it to calculate best possible route.

 

[skydioconvert]

Like they say anything is possible ....

Just don't think the idea will draw a big enough following especially considering what Skydio already has on its plate.

 

[deedeebulldog]

Plenty of outdoor sports have a concept of 'lap' or 'track' or 'trail' in their core - mtb/downhill, alpine skiing, competition enduro racing, motocross etc. it would be beneficial to have a chance to train drone for the lap/track/trail over a set of recorded attempts so once ready, it could deliver super-accurate following of fast-moving object throughout the whole recorded track (because it then know what to expect around itself). As user I'm ready to take consequent runs (or laps) alone same track so that drone could record several attempts and my trajectories and basically take each recorded run to learn and then navigate better through the same terrain. Ideally that after 4-5 practice runs it would be able to deliver flawless tracking without random moves where such maneuvers are not possible.

Imagine I'm riding a mtb downhill trail and I can sacrifice half of a day to make a drone learn the trail: I will do first descend at 10kmh and if/when the drone lost my track I would stop and return it to the trail. Then next run I descend the same trail at 15kmh again making drone learn it at higher speed, then at 20, then at 30, then at full speed. Due to the fact drone has recorded all these attempts with my trajectories and own trajectories over several runs it could basically optimize its own route and probably record 'safe tunnel' for following and flying and as consequence, move faster and better. This mode will help a lot if an object of tracking will move at 30-40-50kmh median speed alone the track and surroundings are not easy to fly through (forest etc). This mode will also so help greatly if I host some event and record each participant attempt - I can pre-practice drone day before and then next day it will capture participants way better than 'out of blind flight following them'

That would be interesting. Like an advanced waypoint mission in a way.

 

[singular.droid]

Like they say anything is possible ....

Just don't think the idea will draw a big enough following especially considering what Skydio already has on its plate.

Yes, this is what I'm afraid most. That this vital for some cases feature will be somewhere in backlog till skydio 9 release