5 ms·

I think AI flown drones is a much more realistic goal now than AI driven cars. It’d certainly make it cheaper and easier for Google and Apple to update maps.

by sykh 8y ago

I think AI flown drones is a much more realistic goal now than AI driven cars. It’d certainly make it cheaper and easier for Google and Apple to update maps. Is there anyone working in this space?

krasin 8y ago
https://www.skydio.com/ https://www.skydio.com/
- sykh 8y ago
  That is cool. If they can pull it off that’d be awesome. Thanks for the link.
  - olex 8y ago
    Skydio R1 is actually on sale and has some seriously impressive tracking and sense-and-avoid functionality going on. FliteTest (a well known RC Youtube channel) has several videos where they test it with several interesting use cases, including high speeds, running between trees, going indoors and otherwise attempting to trick it, and it performs very well, not crashing once. If the next version is a little quieter and smoother on the video tracking, it'll be a serious competitor to manually-flown drones for various sports and "outdoor activity" types of videos.
ballenf 8y ago
Kind of a cool concept to think about construction having unit tests. Autonomous drones that check your work continuously. Prevents a problem from being undetected until after the responsible crew has moved on to the next room/floor/site. But I've probably got it kind of backwards. Manufacturing is where unit tests probably come from and we have been slowly adopting them into programming as we realized their value.
- hobofan 8y ago
  > Kind of a cool concept to think about construction having unit tests. Autonomous drones that check your work continuously. Prevents a problem from being undetected until after the responsible crew has moved on to the next room/floor/site. You have an optimistic view of the digitalization of construction project management. I attended a talk not too long ago from one of the bigger project management solution players in the field and they reported that the status quo for most construction projects is progress tracking via Excel. Structured data in a form that could be used together with autonomous drones is still a few long steps away.
  - nataz 8y ago
    In my experience [1], construction project management econpasses a lot more then just testing, and both levels have to deal with structured and unstructured digital information. There is a lot of Excel and Microsoft project at the higher levels (resource loading, financial tracking), with more specific software tooling for the lower down/more specialized work (think material and labor estimates, load calculations, environmental modeling, other engineering calculations). It's true, a firm building residential track housing may be working off really simple project management tools, but that may be all you need. Start building things like powerplants, bridges, skyscrapers and you'll see a whole new tool set. [1] responsible for $100m+ civil entering and construction projects annually
- acomjean 8y ago
  Construction has lots of tests. At least the landfill construction I was involved with. Each truckload of concrete gets cylinders made. soil/gravel gets sent to labs for testing. Landfill liner welds get field tested and sent to labs to be strength tested for every X ft of welding... I'm not sure how you would use a drone to help this.
excalibur 8y ago
James Cameron.
GFischer 8y ago
At least one Y-Combinator backed startup that I've read of (Iris Automation), and certainly a lot more. https://techcrunch.com/2017/01/18/iris-automation-raises-1-5-million-to-help-drones-sense-and-avoid-obstacles/ https://techcrunch.com/2017/01/18/iris-automation-raises-1-5... Iris Automation has raised money to bring (...) truly self-flying capabilities, to drones used for industrial tasks. Competitors to Iris in the drone industry specifically include SRI spin-out Area 17 (also known as a17), Intel RealSense Technology, Parrot’s SLAMdunk systems and DJI’s Guidance systems.
joshvm 8y ago
Depends what you mean by "AI". Autonomy in aviation is far ahead of what you can get in ground vehicles. Autopilots are commonplace, even in light aircraft. There are established systems like TCAS for traffic avoidance. Aviation remains incredibly safe, despite huge levels of autonomy, because of the regulation involved. What (I think) we need is a regulated system where every drone uses a standard transponder and negotiates airspace either with a controller (e.g. IFR) or other aircraft (e.g. VFR). The future is not everyone flying drones with proprietary AI collision avoidance. One existing product in this space is FLARM which is used by a lot of glider pilots and light aircraft, they also make a UAV variant: https://flarm.com/products/powerflarm/uav/ https://flarm.com/products/powerflarm/uav/ Autonomous drones themselves exist, are inexpensive, and can be set to fly a survey pattern based on a number of pre-selected GPS waypoints. Ardupilot will do this. The data is used for mapping at a much higher level of data than online maps. Think agri, ecology, general site surveying, etc. That said, drones are not ideal for replacing aircraft in the large scale mapping. The flight time is poor. An aircraft modified for surveying can cover hundreds of square kilometres a day. A consumer drone with an hour's flight time (good luck) is going to struggle to compete. Collision avoidance at the local level (e.g. are you going to hit a wall) is a different issue, and that can be solved by standard depth sensors.
- figgis 8y ago
  >That said, drones are not ideal for replacing aircraft in the large scale mapping. You are comparing consumer drones to airplanes. A quick google search tells me you can pick up a drone with over an hour of flight-time for 3.5k-ish (I don't know much about drone pricing so someone may have better options) Another quick google search tells me cheaper planes go from 25k (crappy) to 40k (reasonable) You can purchase 10 drones for the price of one airplane. Which I can't imagine an airplane could out map/survey 10 drones working together. I don't know much about drones or planes though so there are probably other restrictions.
  - abakker 8y ago
    Core difference is /pixel resolution. Drones can easily produce 2-3cm/pixel. Airplanes don't do that well because they can't get that close to the ground and move too fast. For larger areas, though, airplanes are MUCH faster. Realistically, at those larger areas, though, you run against satellites, too.
slavik81 8y ago
For a lot of applications you can already just give it some waypoints and forget about it. What would the AI actually do?

by sykh 8y ago

I think AI flown drones is a much more realistic goal now than AI driven cars. It’d certainly make it cheaper and easier for Google and Apple to update maps. Is there anyone working in this space?

krasin 8y ago

https://www.skydio.com/ https://www.skydio.com/

sykh 8y ago
That is cool. If they can pull it off that’d be awesome. Thanks for the link.
- olex 8y ago
  Skydio R1 is actually on sale and has some seriously impressive tracking and sense-and-avoid functionality going on. FliteTest (a well known RC Youtube channel) has several videos where they test it with several interesting use cases, including high speeds, running between trees, going indoors and otherwise attempting to trick it, and it performs very well, not crashing once. If the next version is a little quieter and smoother on the video tracking, it'll be a serious competitor to manually-flown drones for various sports and "outdoor activity" types of videos.

ballenf 8y ago

Kind of a cool concept to think about construction having unit tests. Autonomous drones that check your work continuously. Prevents a problem from being undetected until after the responsible crew has moved on to the next room/floor/site. But I've probably got it kind of backwards. Manufacturing is where unit tests probably come from and we have been slowly adopting them into programming as we realized their value.

hobofan 8y ago
> Kind of a cool concept to think about construction having unit tests. Autonomous drones that check your work continuously. Prevents a problem from being undetected until after the responsible crew has moved on to the next room/floor/site. You have an optimistic view of the digitalization of construction project management. I attended a talk not too long ago from one of the bigger project management solution players in the field and they reported that the status quo for most construction projects is progress tracking via Excel. Structured data in a form that could be used together with autonomous drones is still a few long steps away.
- nataz 8y ago
  In my experience [1], construction project management econpasses a lot more then just testing, and both levels have to deal with structured and unstructured digital information. There is a lot of Excel and Microsoft project at the higher levels (resource loading, financial tracking), with more specific software tooling for the lower down/more specialized work (think material and labor estimates, load calculations, environmental modeling, other engineering calculations). It's true, a firm building residential track housing may be working off really simple project management tools, but that may be all you need. Start building things like powerplants, bridges, skyscrapers and you'll see a whole new tool set. [1] responsible for $100m+ civil entering and construction projects annually
acomjean 8y ago
Construction has lots of tests. At least the landfill construction I was involved with. Each truckload of concrete gets cylinders made. soil/gravel gets sent to labs for testing. Landfill liner welds get field tested and sent to labs to be strength tested for every X ft of welding... I'm not sure how you would use a drone to help this.

excalibur 8y ago

James Cameron.

GFischer 8y ago

At least one Y-Combinator backed startup that I've read of (Iris Automation), and certainly a lot more. https://techcrunch.com/2017/01/18/iris-automation-raises-1-5-million-to-help-drones-sense-and-avoid-obstacles/ https://techcrunch.com/2017/01/18/iris-automation-raises-1-5... Iris Automation has raised money to bring (...) truly self-flying capabilities, to drones used for industrial tasks. Competitors to Iris in the drone industry specifically include SRI spin-out Area 17 (also known as a17), Intel RealSense Technology, Parrot’s SLAMdunk systems and DJI’s Guidance systems.

joshvm 8y ago

Depends what you mean by "AI". Autonomy in aviation is far ahead of what you can get in ground vehicles. Autopilots are commonplace, even in light aircraft. There are established systems like TCAS for traffic avoidance. Aviation remains incredibly safe, despite huge levels of autonomy, because of the regulation involved. What (I think) we need is a regulated system where every drone uses a standard transponder and negotiates airspace either with a controller (e.g. IFR) or other aircraft (e.g. VFR). The future is not everyone flying drones with proprietary AI collision avoidance. One existing product in this space is FLARM which is used by a lot of glider pilots and light aircraft, they also make a UAV variant: https://flarm.com/products/powerflarm/uav/ https://flarm.com/products/powerflarm/uav/ Autonomous drones themselves exist, are inexpensive, and can be set to fly a survey pattern based on a number of pre-selected GPS waypoints. Ardupilot will do this. The data is used for mapping at a much higher level of data than online maps. Think agri, ecology, general site surveying, etc. That said, drones are not ideal for replacing aircraft in the large scale mapping. The flight time is poor. An aircraft modified for surveying can cover hundreds of square kilometres a day. A consumer drone with an hour's flight time (good luck) is going to struggle to compete. Collision avoidance at the local level (e.g. are you going to hit a wall) is a different issue, and that can be solved by standard depth sensors.

figgis 8y ago
>That said, drones are not ideal for replacing aircraft in the large scale mapping. You are comparing consumer drones to airplanes. A quick google search tells me you can pick up a drone with over an hour of flight-time for 3.5k-ish (I don't know much about drone pricing so someone may have better options) Another quick google search tells me cheaper planes go from 25k (crappy) to 40k (reasonable) You can purchase 10 drones for the price of one airplane. Which I can't imagine an airplane could out map/survey 10 drones working together. I don't know much about drones or planes though so there are probably other restrictions.
- abakker 8y ago
  Core difference is /pixel resolution. Drones can easily produce 2-3cm/pixel. Airplanes don't do that well because they can't get that close to the ground and move too fast. For larger areas, though, airplanes are MUCH faster. Realistically, at those larger areas, though, you run against satellites, too.

slavik81 8y ago

For a lot of applications you can already just give it some waypoints and forget about it. What would the AI actually do?