Back
[00:18:06] <Triffid_Hunter> cyberdelic: no sensor is accurate in all conditions. sonar can't see soft or thin things and certain objects like tennis balls, IR can't see anything that doesn't diffusely reflect IR (eg green things, garbage bags, mirrors, windows), LIDAR is basically glorified high speed IR with a spinner
[00:18:21] <Triffid_Hunter> cyberdelic: your best bet is to use multiple sensors and perform sensor fusion on the results
[00:18:36] <orlock> hey Triffid_Hunter
[00:18:53] <orlock> hows it going in the land of robots and counterfiets?
[00:19:08] <Triffid_Hunter> orlock: good, narrowing down on a working SLAM implementation at the moment
[00:19:33] <orlock> for a next-gen toy?
[00:20:10] <orlock> i finally put in an upgraded X carridge and extruder in my prusa, seems to have sorted out some of my print issues
[00:20:11] <Triffid_Hunter> yup
[00:20:28] <orlock> works having a sumo challenge
[00:20:51] <orlock> 3kg 20cm/sq or 10cm/sq 300g
[00:21:17] <orlock> going to revive my old chopping board bot
[00:23:12] <orlock> i wonder if PLA gears will stand up to it
[00:27:12] <cyberdelic> cheers Triffid_Hunter :)
[00:36:44] <MrCurious> so green tennis balls are the nemesis of robot sensor systems...
[00:37:16] <Triffid_Hunter> MrCurious: heh yep, use computer vision for those ;)
[00:37:33] <MrCurious> thats what i was thinking cv
[00:38:00] <Triffid_Hunter> so easy to slap cv on robots these days, what with stuff like the odroid c1 and similar small ARM boards widely available for peanuts
[00:38:58] <MrCurious> or tegra tk1's available for $100 if you acted at the right moment
[00:39:11] <MrCurious> or $127 for intel nuc's centrino's
[00:39:29] <MrCurious> yeap, the age of disposable computing is upon us
[00:39:34] <Triffid_Hunter> tegras are awesome, but odroid c1 is adequate and $35 :P
[00:39:55] <Triffid_Hunter> way better than RPi, it has two usb root ports and real ethernet
[00:40:25] <MrCurious> i am thinking the 192 cuda cores will have their place
[00:40:45] <Wetmelon> what is this odroid you speak of
[00:40:48] <MrCurious> got a realsense r200 last week and a pair of nuc's (centrino, i3)
[00:41:03] <Triffid_Hunter> sure if you're doing deep learning, but finding green circles doesn't require nearly that much oomph ;)
[00:41:36] <Wetmelon> Triffid_Hunter: What OS does this Odroid run? Android?
[00:41:45] <Triffid_Hunter> Wetmelon: odroid C1+, about 2x faster than RPi's CPU, two USB root ports (important for hooking up multiple cameras) and real gigabit ethernet that goes to an on-SoC MAC
[00:41:51] <Triffid_Hunter> Wetmelon: android or ubuntu
[00:42:01] <Wetmelon> Neat!
[00:42:13] <Wetmelon> I'm trying to work out the embedded system for a race car
[00:42:54] <MrCurious> Asus i3 nuc + (3) playstation eye webcam. 320x240 @ 187FPS
[00:42:58] <Wetmelon> Not sure what blend of bare metal MCUs, FPGAs, and more powerful non-RTOS stuff I need
[00:43:50] <Triffid_Hunter> Wetmelon: autonomous race car? not sure you'll get nearly enough performance from an odroid for that, we'll soon be running SLAM at ~5fps at the moment which is adequate for our application
[00:44:20] <MrCurious> Triffid_Hunter: got a pic of the bot or a spec sheet?
[00:44:24] <Triffid_Hunter> MrCurious: yeah I hear those ps eye cams are awesome, can't find 'em for reasonable cost anywhere though, esp now that they're apparently discontinued
[00:44:39] <MrCurious> http://www.ebay.com/itm/Playstation-3-Official-Eye-USB-Camera-Brand-New-/111816831178?hash=item1a08cd70ca:g:UpIAAOSw7hRWO3gF
[00:44:44] <Triffid_Hunter> MrCurious: for mine? nope, it's commercial and still under wraps
[00:44:46] <MrCurious> $4.50 ea seems reasonable to me
[00:44:54] <MrCurious> Triffid_Hunter: check your PM
[00:44:57] <Triffid_Hunter> MrCurious: that's 4x what we're paying for cameras here :P
[00:45:08] <Wetmelon> nah, driven car.
[00:46:06] <Wetmelon> you working with SLAM? I wish is known that earlier. One of our senior project groups is building a SLAM bot
[00:46:13] <Triffid_Hunter> Wetmelon: oh full size? use a couple of i5s and a bunch of arduinos for hardware interfacing :P
[00:48:40] <Wetmelon> Triffid_Hunter: Wat. i5's are WAY overkill for what we need haha. The only OS stuff I need is data acquisition / telemetry.
[00:48:51] <Wetmelon> I really just need something that can pick up a CAN bus and transmit it to the ground station
[00:49:15] <Wetmelon> Planning on mostly using Teensy dev boards (Arduino compatible ARM core)
[00:49:20] <Triffid_Hunter> Wetmelon: oh, plenty of MCUs with CAN these days.. I thought you were doing SLAM with lidar and cameras and whatnot
[00:49:30] <Wetmelon> Nooo, that's the other senior project team.
[00:49:37] <Wetmelon> They're using a Project Tango and some other stuff
[00:50:38] <Wetmelon> This is just a small electric single-seat electric race car.
[00:51:03] <Triffid_Hunter> Wetmelon: oh right, not autonomous?
[00:51:14] <Wetmelon> Nah
[00:51:16] <Triffid_Hunter> Wetmelon: using a big BLDC or a series wound DC motor?
[00:51:37] <Wetmelon> Dual PMAC is the plan if we can afford it. Dual series wound DC if we can't :P
[00:52:19] <Wetmelon> On a 600lb car it should be fun.
[00:52:20] <Triffid_Hunter> Wetmelon: heh pmac is a better name than bldc
[00:52:56] <Triffid_Hunter> Wetmelon: I've read that the main reason they're not using pmac in teslas and similar is switching losses and EMI are astronomical :/
[00:53:50] <Wetmelon> I thought it was mostly so they didn't have to find huge magnets. They use 3-phase AC induction iirc?
[00:55:02] <Wetmelon> The switching losses are... impressive. We're limited to 80kW but even that is beyond the range of any sort of DC/DC converter, and outside most COTS motor controllers
[00:55:41] <Wetmelon> 95% efficient DC/DC? Great! That's only 4kW of loss! Hah
[00:56:05] <Triffid_Hunter> oh they're induction motors? interesting, I thought they were pmac
[00:56:12] <Triffid_Hunter> I mean series wound, damnit
[00:59:00] <Wetmelon> It's okay I gotchu :P
[01:00:07] <Triffid_Hunter> I saw a teardown of the controller in the tesla roadster, they had IGBTs bigger than postage stamps in there
[01:03:37] <Wetmelon> The Model S now has an upgrade available (for the ludicrous acceleration option) that has to replace the 1300A main fuse with a 1500A inconel main fuse... how the fuck do you switch 1500A!?
[01:04:45] <Triffid_Hunter> Wetmelon: big IGBT
[01:05:02] <Triffid_Hunter> Wetmelon: no seriously, those things are crazy. also it's not just 1500A, but 1500A at like 500v
[01:05:45] <Wetmelon> Off-topic, but I got to drive a Roadster a week or so ago :D
[01:05:49] <Triffid_Hunter> Wetmelon: beyond 100A IGBTs start becoming more attractive than mosfets since I²R starts to exceed I*Vce(sat)
[01:06:00] <Wetmelon> Right
[01:06:12] <Triffid_Hunter> even though Vce(sat) for IGBTs is as much as a couple of volts, that I² term is a killer for high currents
[01:06:13] <Wetmelon> Do you have to do anything fancy to drive IGBTs?
[01:06:41] <Triffid_Hunter> Wetmelon: input is basically the same as a mosfet, however due to the ridiculous currents and dv/dt the big ones are often optically triggered
[01:06:46] <Wetmelon> Or just what you'd usually use for high power (high gate capacitance) mosfets
[01:07:03] <Wetmelon> Ah
[01:07:23] <Wetmelon> I'm halfway tempted to make my own motor controllers for the race car >.>
[01:08:21] <Triffid_Hunter> Wetmelon: would certainly be a fun project.. fun as in lots of fire from lots of large expensive components ;)
[01:08:54] <Wetmelon> xD
[01:09:19] <MrCurious> ls
[01:19:35] <Wetmelon> How do I decide how many sensors / peripherals etc to put on a single MCU?
[01:20:16] <Wetmelon> Should I use one processor as much as possible, or really limit the wiring distance and lean more heavily on the CAN bus?
[01:25:27] <Triffid_Hunter> Wetmelon: group sensors that are physically close (within ~20cm or so), use CAN for longer runs
[01:31:10] <Wetmelon> That gives me a rough range, thanks.
[01:40:45] <rue_shop3> ok forearm rotator motor is working
[01:43:39] <rue_shop3> need to work out getting 6 drive cables down a 3/4" pipe
[02:03:25] <rue_house> breaking them out on the wrist end is the hard part I think
[02:06:03] <SpeedEvil> https://upload.wikimedia.org/wikipedia/commons/f/f0/Carpal-Tunnel.svg
[02:12:54] <MrCurious> naproxin :)
[02:13:13] <MrCurious> will work for a while... then you have to give up and get the surgery
[02:29:56] <rue_house> all 6 in a 3/4" pipe
[02:30:14] <rue_house> 2 for the wrist tip, 2 for wrist twist, and 2 for the gripper
[02:30:55] <rue_house> I'm debating putting the gripper motor on the gripper tho
[09:29:38] <rue_house> shower logistics
[09:30:42] <rue_house> its easier to build a Zrotate-Xtip-Ytip-grip arm than it is to build a Zrotate-Xtip-Zrotate-grip arm
[09:38:58] <rue_house> armatron is only 6 axies cause thats all they could do on the sticks
[09:39:02] <rue_house> hmmm
[16:56:19] <orlock> so, i think if make two sets of 90 degree bevel gears, i can fit my existing robot parts into a 20cm square, just
[16:56:44] <orlock> 16cm long motors, 5cm wide
[16:57:20] <Snert__> I'd like to see that.
