#robotics Logs

May 10 2015

02:11 al-caveman how to know which brushless dc motor is good for me to make a drone out of it?
02:16 rue_shop2 get a drone and find out what it uses
02:17 al-caveman that's too specific. i want a more general criteria
02:18 al-caveman perhaps i should sample 1000s of drones and then generalize a criteria of what works for drones and what not but this is too costy and not smart
02:28 deshipu al-caveman: all depends on what kind of drone it is. A wheeled drone? A tracked drone? Walking? Underwater?
02:31 al-caveman deshipu: flying drone
02:34 deshipu al-caveman: pushing or pulling propeller?
02:34 deshipu al-caveman: what wing configuration?
02:34 al-caveman dunno what's pushing/pulling i think they are all the same just suck air
02:35 deshipu I see, ok, I have no more questions.
02:35 al-caveman no wings. just a quad copter
02:35 deshipu are you sure you want it to be ad drone, and not just remote control it?
02:36 al-caveman ad drone?
02:36 deshipu a drone
02:36 deshipu sorry
02:36 al-caveman i may remote control it too, or put some sensory on board and let AI do the thing
02:37 deshipu ah, that should be easy, right?
02:37 al-caveman the AI bit or the remote control?
02:39 deshipu the ai bit
02:40 al-caveman yeah easy although lots of fun and experiments but the path is open. can do statistical models for it too in case deterministic ones don't suffice
02:42 deshipu al-caveman: why don't you start with one of those cheap chinese toys, and once you know more about them, start building your own?
02:43 deshipu al-caveman: I remember that someone actually replaced the firmware to open sourc one in those small hubsans
02:43 deshipu those $20 ones
02:44 rue_shop2 hmm 3 robot arms
02:45 al-caveman deshipu: cause not fun. i want to enjoy building one
02:45 al-caveman deshipu: i don't enjoy being a consumer in this part of the market
02:46 al-caveman but i'm okay if i'm a consumer for motors or basic stuff. these save time and not much fun wiring a motor
02:46 deshipu al-caveman: well, you will enjoy hacking it
02:46 deshipu al-caveman: and programming, and adding sensors
02:46 al-caveman nope. not really fun to reverse engineer the works of a chinese
02:46 deshipu al-caveman: basically you will skip the frustrating part and skip to the fun part
02:47 al-caveman thing is, i can't add my own stuff easily because of decisions made by the chinese engineer
02:47 deshipu al-caveman: why not?
02:47 al-caveman maybe the chinese decided some go with some silly micro controller with silly code that is not capable of integrating nicely with my AI controllers
02:47 deshipu that french guy doing excess flood all the time starts to annoy me
02:48 rue_shop2 oops
02:48 al-caveman i'm also starting dealing with micro controllers for the 1st time. i want to do my own creation. i feel like i love this
02:48 rue_shop2 one robot arm is broken
02:48 deshipu rue_shop2: you broke it? how coud you!
02:48 deshipu could
02:49 rue_shop2 another robot arm fell on it twice, I'm not tooo surprised
02:50 deshipu al-caveman: ah, the good old "I have no idea what I'm doing, so I want to make all my decisions myself instead of trusting people with experience" thing
02:50 al-caveman maybe
02:50 deshipu well, good news is that in that case it doesn't matter what motor you choose
02:51 deshipu you may as well pick the cheapest one, as it won't fly anyways
02:51 deshipu you may even skip the motors, the result will be the same
02:51 al-caveman is this channel open to tell people to buy existing toys made by china?
02:52 rue_shop2 the two robot controllers work great
02:52 deshipu rue_shop2: make them play rock paper scissors
02:52 rue_shop2 but the robots have different axis characteristics
02:52 rue_shop2 all the zeros are in different places
02:52 rue_shop2 the scaling is different
02:52 rue_shop2 and the axies are reversed in some cases
02:53 deshipu would be boring to make two idenitical arms, no?
02:53 rue_shop2 so, I kinda need a way to know which arm I'm talking to
02:53 deshipu embed version and model in the handshake/
02:54 deshipu no, wait
02:54 deshipu there is no handshake, is there?
03:00 deshipu rue_shop2: how about a "key" with different jumper settings?
03:00 deshipu like, a piece of plastic with the jumper connections empedded, hanging on a string from the robot
03:01 deshipu of course some kind of memory with all the parameters and configuration would be nicer
10:46 deshipu yay
10:47 deshipu got my robot to switch from creep to trot automatically when more speed is needed
10:48 slidercrank http://www.youtube.com/watch?v=EOzGQMnRTNw Incredible: Robot Stitches A Grape Back Together!
10:49 deshipu slidercrank: I'm not sure if it's actually an autonomous robot...
10:49 slidercrank I think there is an operator controlling it remotely but still
10:50 deshipu consider how much more work it would be if it had to analyze the scene, plan the movements, etc.
10:50 slidercrank https://en.wikipedia.org/wiki/Da_Vinci_Surgical_System
10:51 slidercrank The da Vinci System consists of a surgeon’s console ..
10:51 deshipu lame :)
10:53 deshipu I still can't believe how well that robot I liked yesterday walks
10:53 veverak :D
10:53 slidercrank deshipu, what robot? link?
10:54 deshipu https://www.youtube.com/watch?v=-Vg-BdXps50
10:55 deshipu this is so much better than the robots from the DARPA challenge
10:57 veverak given the amount of resources we invested into walking robots
10:57 veverak I sometimes wonder ... why is it so hard to make properly walking thing ?
10:58 deshipu veverak: because proper walk is very dynamic, and that involves differential equations that we can't solve in closed form
10:58 veverak I see
10:58 deshipu veverak: so we have to simplify them and/or solve them numerically, using iterating optimizations and so on
10:59 deshipu veverak: they actually do have a good model of a humanoid robot and they can simulate it walking
10:59 deshipu veverak: on a modern computer simulating 30s takes a day
10:59 veverak I see
10:59 deshipu so there are three solutions, and we use a mix of them
11:00 deshipu 1. simplify the model, 2. use tricks to find good enough solutions, 3. more computing power
11:00 verak should learn more about these equations
11:00 deshipu I mostly do the extreme form of 1), that guy from this link mostly did 2)
11:01 deshipu veverak: that underactuated robotics course is awesome for that
11:01 veverak bookmarked :)
11:02 veverak I always suspected it was problem of mechanics... but nah, my bad :)
11:04 deshipu veverak: https://www.youtube.com/watch?v=NAkfl3kKnVA
11:04 deshipu veverak: this is where he gives the most useful information
11:05 deshipu I'm not sure if you can access the course itself: https://courses.edx.org/courses/MITx/6.832x/3T2014
11:06 veverak thanks! :)
11:07 veverak nah, have to find proper material on D*
11:07 veverak it looks pretty amasing
11:10 deshipu looks like there isn't that much about it around
11:10 veverak yeah
11:11 veverak hmm
11:11 veverak I have now subject "algorithms"
11:12 veverak maybe they will be able to point me to something
11:12 veverak did graph algorithms las time
11:13 veverak but only Bellman-Ford and Dijkstra
11:14 veverak but nah, I want something that is able to work with "graph" that changes :)
11:22 deshipu veverak: no Prime and Kruskal? :D
11:24 veverak nope
11:30 deshipu veverak: they can be useful, I used them for generating mazes
11:31 verak likes the subject,
11:31 veverak I don't want to see redblack trees for a while
11:47 deshipu wow, the Japanese really are ahead: https://www.youtube.com/watch?v=Qqz28b6SLzQ
11:50 zaxxon_ https://www.youtube.com/watch?v=FxM40qBK02U , damn
12:01 deshipu I wonder what sensors it has
12:01 veverak nah
12:01 veverak which reminds me another todo
12:01 deshipu probably pressure sensors in the feet
12:01 veverak two webcams -> 3d
12:02 deshipu does opencv have it?
12:02 veverak dunno
12:02 veverak I am just intersted in the topic
12:02 veverak more like interested in a ways about how to scand surroundings of robot
12:02 deshipu I suppose the hardest part is finding the corresponding points on both cameras
12:03 veverak and multiple webcams -> 3d seems interesting
12:03 veverak exactly
12:03 veverak :)
12:03 veverak but I guess some of the algorithms "I don't KNOW what is there, but I bet it's 80% chance it's like this..."
12:03 veverak you know what I mean...
12:04 deshipu nah, even humans don't do that with stuff like necker cube
12:04 deshipu we just choose one option randomly
12:04 veverak yeah :)
12:08 veverak deshipu: I just suppose that it will be algorithm base on the principle
12:08 veverak "there is 80% chance that this two points are the same, which means they are on the position of XY"
12:09 veverak sum it together, there is 76% chance that on that coord, is solid thing
12:09 veverak on the other hand, there is 15% chance there is something next to it -> go around it
12:16 LiohAu deshipu: It's really hard to find a ball joint
12:18 LiohAu I think that I should forget the 3rd rotation axis, keeping only pan/tilt so I can use a low cost pan tilt tower
12:37 deshipu LiohAu: yeah, humans rarely tilt their heads anyways
12:37 deshipu veverak: add time and it all becomes 1000% more complex
12:45 veverak I know
13:36 rue_house wholy geez, there is an isp that is 1.0.0.0 / 8
13:37 rue_house thats 1/256 of the ipv4
13:39 rue_house er, I think I did that wrong
16:08 rue_shop3 ok I have 3 arms i need to do soemthing fora fair on the 31st
16:16 SpeedEvil rue_shop3: freak show?
16:17 SpeedEvil Juggling? Plaiting? Knitting? Knife-throwing?
16:33 deshipu fencing
16:50 SpeedEvil yeah - I've got some posts I need to install outside.
16:51 SpeedEvil I just need to build an arm to push them in
16:51 SpeedEvil ~4 or 5 tons should work
17:58 rue_shop3 just maker fair
18:05 veverak are there any known algorithms
18:05 veverak to get 3D .. 'idea' of actuale surroundings from traditional cameras?
18:05 veverak let's say 1080p webcameras
18:06 veverak I have two webcameras, I know thery position relative to each other
18:06 veverak is there some known way about how to get 3D data from that setup?
19:31 Krypteron o/