PDA

View Full Version : Maxwell: A Mobile Manipulator



lnxfergy
02-06-2011, 08:23 PM
I wanted to briefly introduce the robot I've been working on for the past couple weeks. Maxwell is a human-scale mobile manipulator powered by ROS and an ArbotiX.

http://forums.trossenrobotics.com/gallery/files/1/7/6/8/sm_maxwell.jpghttp://forums.trossenrobotics.com/gallery/files/1/7/6/8/rsm_maxwell.jpg

A cool feature of this bot is that not only does it sport lots of sensory and a cool arm, but Maxwell breaks down into short sections that fit in a Pelican case. It only takes about 5 minutes to break down or assemble the robot:

http://forums.trossenrobotics.com/gallery/files/1/7/6/8/sm_packed.jpg

I've posted more pictures and details on my blog: http://www.showusyoursensors.com/2011/02/introducing-maxwell.html for now, I'll expand this post shortly.

-Fergs

Zenta
02-07-2011, 05:46 AM
You never stop to impress me with all your projects.
This one looks very interesting.
You say that the wrist has a roll function too? It was very hard to see it on the pictures though.

Looking forward see your next post.

elaughlin
02-07-2011, 06:15 AM
Is the arm made up of mostly RX servos? And then a couple of AX's at the gripper? Must have some good stability in the arm from the upgraded servos.

Amp
02-07-2011, 01:41 PM
Looks really cool, and some very very nice servos :-)
How hard is it to get ROS working with your own gripper?
What library will you/ can you use for that?

lnxfergy
02-07-2011, 08:41 PM
You never stop to impress me with all your projects.
This one looks very interesting.
You say that the wrist has a roll function too? It was very hard to see it on the pictures though.

Looking forward see your next post.

Yep, there is an AX12 right behind the 2 AX-12s used for the gripper fingers. It allows the wrist to rotate to grasp an object from different angles. I'll try to get a better picture up soon, but you can sorta see the top side of the servo in this pic (the horn of the AX-12 is connected to the RX-64 hinge bracket):

http://3.bp.blogspot.com/_qmthA_fNJ-Y/TU9KY9oGmRI/AAAAAAAAAM8/PWlCuKTJRVY/s1600/sm_arm.jpg


Is the arm made up of mostly RX servos? And then a couple of AX's at the gripper? Must have some good stability in the arm from the upgraded servos.

Yep, my blog post had a bit more detail: "Maxwell's arm is constructed from 2 EX-106 (shoulder lift and elbow flex), 2RX-64 (shoulder pan and wrist flex), and and 3 AX-12s (one for wrist roll, and two to form a gripper). In all honesty, it only needs the one EX-106 in the shoulder, but I didn't have any brackets that fit the RX-64 at the moment. The actual gripper fingers are temporary, I have much better ones in the works."


Looks really cool, and some very very nice servos :-)
How hard is it to get ROS working with your own gripper?
What library will you/ can you use for that?

Thanks. I'm not going to lie: getting a custom manipulator up and running in ROS is a bit of a task. I'm using the arm_kinematics package to do IK -- with my own wrapper around it do deal with the 5DOF-ness of the arm. I'm then using the collision_environment to apply collision free kinematics (still in progress), and tying it all together with my own simple version of move_arm (eventually I'd like to use the actual move_arm -- but it currently has some PR2 dependencies I'm avoiding).

The good news is: I firmly believe in open source. All of the trajectory planning aspects will be integrated into upcoming ArbotiX package releases for ROS. This should make it fairly straight forward to create a custom URDF+YAML description of your AX/RX servo arm, connect it to an ArbotiX and go forward with arm planning. Maxwell's description files will also be posted shortly (some already are, but it's somewhat of a mess, I was working rapidly over the weekend to get him operational for a full day of demos today).

-Fergs

RobotAtlas
02-07-2011, 09:37 PM
constructed from 2 EX-106 (shoulder lift and elbow flex), 2RX-64 (shoulder pan and wrist flex), and and 3 AX-12s
Multi-servo-type approach makes a lot of sense to me, especially when I look at my own arm which has one of the biggest muscles in the body to move arm(proper) and so much smaller muscles to move my fingers.

This setup still a lot more affordable than $160k(?) price tag of Barrett arm too. :)

Too bad ArbotiX can't do RX and AX at the same time...

lnxfergy
02-07-2011, 09:45 PM
Too bad ArbotiX can't do RX and AX at the same time...

Stick around..... there's a super secret prototype in the base of Maxwell....

-Fergs

lnxfergy
02-11-2011, 04:52 PM
http://www.youtube.com/watch?v=SDk41lxjcfI

-Fergs

maitri982
02-13-2011, 02:44 PM
You have done a great job with Maxwell. There could be commercial potential for such a robot.

May I ask what material the base is made of?

Also, are you concerned with the height of the arm that the robot, even having a low center of gravity, will easily tip over when the arm is grabbing an item?

Kevin

lnxfergy
02-13-2011, 03:04 PM
You have done a great job with Maxwell. There could be commercial potential for such a robot.
Thanks.


May I ask what material the base is made of?It is laser cut ABS, mostly 3/16" thick but the top piece is 1/4" thick.


Also, are you concerned with the height of the arm that the robot, even having a low center of gravity, will easily tip over when the arm is grabbing an item?Not really -- Maxwell is about 15lbs fully loaded, 10 pounds of that is in the base. His rear end may tip up just a bit, but by keeping the arm speed slow, he doesn't get enough momentum to carry all the way to a tip over.

He probably won't be able to negotiate ramps as well as something like a PR2 -- but that is not an issue in the environments I'm working in currently. Maxwell has successfully negotiated a number of very large door and carpet transition thresholds without incident.

-Fergs

mbahr
02-13-2011, 03:48 PM
I have an idea for a robot that could also use a vertical sensor riser as you have shown. Pretty nice!

So the top is 1/4" ABS?

Is the sensor riser you have 1.5"x1.5" 80/20 or bosch aluminum framing? How is it anchored? Do you think the 1/4" ABS is strong enough to avoid long-term stress cracks from the anchor?

-Mike

Pi Robot
02-13-2011, 06:06 PM
Hey Fergs,

The more I look at Maxwell, the more I am inspired to imitate. One feature I would find desirable in a fully autonomous robot is the ability to self-recharge. Rather than reinvent the wheel, that would mean using something like a Create, Neato, or even a Rovio as the base. I know you have extensive experience with the Create and Neato and I think the Rovio footprint is too small to support something like Maxwell's height, so if you had to choose one of these for a base for Maxwell which would it be? Which would give you the best odometry for ROS navigation? And would the Neato's laser be essentially blocked by the sensor riser? Or can you run the Neato backwards instead of forwards (assuming the laser has a 360 degree range)?

--patrick

lnxfergy
02-13-2011, 06:51 PM
Is the sensor riser you have 1.5"x1.5" 80/20 or bosch aluminum framing? How is it anchored? Do you think the 1/4" ABS is strong enough to avoid long-term stress cracks from the anchor?

The column is 8020, profile 1010 (1"x1"). It's anchored to the top plate by 2 right angle brackets and slips into a "socket" on top of the bottom plate. I doubt there will be issues with this mounting.

-Fergs

lnxfergy
02-13-2011, 07:01 PM
The more I look at Maxwell, the more I am inspired to imitate.

Imitation apparently is the most sincere form of flattery -- so thank you!


One feature I would find desirable in a fully autonomous robot is the ability to self-recharge. Rather than reinvent the wheel, that would mean using something like a Create, Neato, or even a Rovio as the base.I totally agree. At the moment Maxwell is mostly a research platform and I'm not expecting him to run fully autonomously, however I do have plans for such down the road. I'm on a bit of a tight time table to get him functional for the AAAI challenge and skipped a few of his additional features at the moment (such as the vertical lift on the arm and the recharging aspects).


I know you have extensive experience with the Create and Neato and I think the Rovio footprint is too small to support something like Maxwell's height, so if you had to choose one of these for a base for Maxwell which would it be? Which would give you the best odometry for ROS navigation? I'm honestly not sure that either of the Create nor Neato is large enough to carry something the size of Maxwell. The wheel arrangement on both means they are at least 25% smaller footprint at the point of contact with the floor. I haven't tried a large boom on the Neato, but on the Create I wouldn't dare go as tall as Maxwell. Nelson was just under 3' tall, and was very tipsy (while still being a bit lighter than Maxwell's arm + head) -- and that was with an extra 5lb battery in the base as added ballast.

Odometry is comparable -- although I think the Neato has the upper hand by a bit. Base control however is better on the Create (as the laser data capture consumes a lot of usb bus time and you are limited to 5hz update rate on the base control as a result -- this is still sufficient for the nav stack as far as I've seen).


And would the Neato's laser be essentially blocked by the sensor riser? Or can you run the Neato backwards instead of forwards (assuming the laser has a 360 degree range)?You'd probably want to run the robot backwards to avoid really trashing the laser view as the dustbin is one of the few places you could easily attach an upright column. The "running backwards" mode actually makes the laser data better in ROS terms, as it would work out of the box with gmapping and such due to where the scan begins. We'll have new launch files for doing this in the next few days -- I'm currently working to prepare an 0.1.0 release of neato_robot so that we can begin building debs within the ROS release framework.

-Fergs

Pi Robot
02-13-2011, 07:30 PM
Thanks Fergs--all very cool. I wonder if one might just add "outrigger" wheels to the base to help with the balance. Of course, they would then become a tripping hazard and could get caught on corners and other objects...So probably a non-starter.

Another possibility might be to retrofit the iRobot charger mechanism onto your own base. Does the charger portion unbolt easily from the robot?

There's a Frys Electronics just down the road with a Neato waiting on the shelf, but I already have this super nice Hokuyo laser scanner (as I know you do) and Willow Garage seems to be focusing on the Create for its turtlebot. As you may have seen in the HBRC videos, the turtlebot uses a laptop slid onto the robot as I believe you do with Maxwell. I wonder how much added stability that extra weight provides?

--patrick

maitri982
02-13-2011, 07:38 PM
Sorry for pummeling you with questions Fergs...but when you do something this interesting....

One comment is that it would be good to get a complete spec on the robot. What wheel motors did you use? What is the wheel diameter? etc etc. Even some pics of wiring. I can tell you this would likely help sell your board as many would like to imitate portions of what you have done here.

I had a question about your future plan to put a LIDR on the base. Why not use a Kinect there as well and do what the willow garage folks are doing and basically taking a plane of its input to simulate a laser plane scan? Is it because of the Kinects up close blind spot?

thanks for responding to these queries and I appreciate your inventiveness!

Kevin

lnxfergy
02-13-2011, 07:55 PM
One comment is that it would be good to get a complete spec on the robot. What wheel motors did you use? What is the wheel diameter? etc etc. Even some pics of wiring. I can tell you this would likely help sell your board as many would like to imitate portions of what you have done here.

Eventually I will have all this posted -- Maxwell is literally a week and a half old right now (the base parts only arrived last Friday) and I'm on the tail end of a huge rewrite of the ArbotiX ROS wrappers. My current priorities are on getting the arm IK refined (almost done), getting the Kinect calibrated to the base (in progress), and getting move_arm up and running a collision environment. Once I've got grasping working, I'll return to document things more closely. For now, there are a few more pictures on my blog post of the inside of the base...


I had a question about your future plan to put a LIDR on the base. Why not use a Kinect there as well and do what the willow garage folks are doing and basically taking a plane of its input to simulate a laser plane scan? Is it because of the Kinects up close blind spot?I'm more concerned with the narrow angle of the Kinect. It's really just the wrong tool for the job. I have on-hand a nice Hokuyo, that was designed for mapping and localization. Getting mapping/localization working with a Kinect means eating up a lot of extra processing power that could be better spent on other things (not to mention the difference in energy consumption between a Hokuyo or Kinect+PC).

On a side note, I also haven't actually seen anyone do localization off just the point_cloud_to_laserscan output of the Kinect stuff. I'm very skeptical that it will work reliably (localization from 3d features is do-able, but again, at a much higher processing cost).

My focus right now really is a reliable, human-scale, mobile manipulator for under $5k (and preferably around $3k). Maxwell's parts are currently at about $3400 + laptop (including a $1200 laser).

-Fergs

lnxfergy
02-13-2011, 08:02 PM
Another possibility might be to retrofit the iRobot charger mechanism onto your own base. Does the charger portion unbolt easily from the robot?

I don't think so -- I believe it's built into the bottom of the base on the Create.


There's a Frys Electronics just down the road with a Neato waiting on the shelf, but I already have this super nice Hokuyo laser scanner (as I know you do) and Willow Garage seems to be focusing on the Create for its turtlebot. As you may have seen in the HBRC videos, the turtlebot uses a laptop slid onto the robot as I believe you do with Maxwell. I wonder how much added stability that extra weight provides?

The laptop certainly helps -- although Nelson also had a netbook on the rear deck as well. The big problem with the Create is that the suspension is "squishy", and so it allows a huge amount of rocking back and forth --- which quickly turns into oscillation when there is any weight up high on a mast.

-Fergs

Pi Robot
02-13-2011, 08:11 PM
The big problem with the Create is that the suspension is "squishy", and so it allows a huge amount of rocking back and forth --- which quickly turns into oscillation when there is any weight up high on a mast.
-Fergs

I know what you mean--we have one of the early Roomba vacuum cleaners. But I was hoping they firmed things up for the Create...

Good luck with the move_arm stuff. I'm waiting to learn from what you come up with. I just installed your arbotix 0.4.0 release and rosmake went through without a hitch. So far so good!

--patrick

lnxfergy
02-13-2011, 08:15 PM
Good luck with the move_arm stuff. I'm waiting to learn from what you come up with. I just installed your arbotix 0.4.0 release and rosmake went through without a hitch. So far so good!

I'm shooting to have 0.4.1 out tomorrow or Tuesday, with updated controllers (now with interpolation) and a first pass at diagnostics tie in -- so let me know if you encounter anything and we can fix it for the next release.

-Fergs

mbahr
02-14-2011, 05:56 PM
I found this the other day and thought it may be useful for inspiration for charging Maxwell.

http://www.superdroidrobots.com/shop/item.aspx?itemid=1042

-Mike

Pi Robot
02-15-2011, 07:05 AM
Not really -- Maxwell is about 15lbs fully loaded, 10 pounds of that is in the base.
-Fergs

Hey Fergs,

Would you mind saying what motors and encoders you are using with Maxwell? At some point I'd like to move my drive and odometry over from the Serializer to the ArbotiX so I might as well match what you are doing.

Thanks!
patrick

lnxfergy
02-15-2011, 08:56 AM
Hey Fergs,

Would you mind saying what motors and encoders you are using with Maxwell? At some point I'd like to move my drive and odometry over from the Serializer to the ArbotiX so I might as well match what you are doing.

Thanks!
patrick

Patrick,

Maxwell's motors and caster are from one of these http://www.zagrosrobotics.com/shop/item.aspx?itemid=529. I'm actually using the 2x speed model, which seems to be OK with this level of payload.

I replaced the drive wheels with BaneBots 4-7/8" wheels, giving a top speed of around 0.5m/s. Currently I'm using 6mm hubs that have been bored out to 1/4" -- although I've now got some parts here to try out the larger 1/2" hubs with a reducer to 1/4" (so that the set screw is a much larger size).

I'm using the same Pololu motor drivers.

-Fergs

Pi Robot
02-15-2011, 09:18 AM
Very nice! Thanks for the link. I don't think I've ever been on that site before. Don't know how I missed it.

--patrick

lnxfergy
03-20-2011, 07:43 PM
A few updates on Maxwell. He was rebuild this weekend to install an updated version of the base (which looks identical to the untrained eye, but is infinitely easier to access the innards of), install an emergency stop switch (not sure how that was forgotten..), and clean up the wiring.

I've posted the current iteration of code and configuration files up on a new repository: fergy-ros-pkg. It's hosted at Google Code like my other repositories. The wiki contains a couple pages of notes.

I've also mostly finished getting the navigation stack tuned on Maxwell. He is creating quite nice maps, and AMCL is working well with the maps. There is still a bit of "hallway shortening" going on, but the higher mounting position of the Hokuyo laser has helped a lot. (http://www.showusyoursensors.com/2011/03/sensors-hate-dark-surfaces.html) I've finally been able to build a complete map of our labs:

http://1.bp.blogspot.com/-ynVnrwgnY7M/TYVpdi-PzFI/AAAAAAAAANs/lVH_9gkRSeQ/s320/ils.png

I should have some videos up later this week of Maxwell navigating the lab. Just putting finishing touches on the semantic map overlay and voice control to move from waypoint to waypoint....

Lastly, there was a request for a parts list -- it's been up on the wiki for some time now: parts list (https://spreadsheets.google.com/pub?key=0AoEGbFgI5kOhdG9fRkFTVnc5VXkyZ0JYcW5laDE3N kE&hl=en&output=html). Unfortunately, it's already a bit out of date. Should be updated shortly.

-Fergs

lnxfergy
03-30-2011, 02:27 PM
http://www.youtube.com/watch?v=RtdCnI0Kf6U

Very early version of moving a chess piece.

-Fergs

Pi Robot
04-06-2011, 11:10 AM
Hey Fergs,

This is a fantastic result! I've barely begun trying to get arm IK working on Pi Robot. Can you say which kinematics package you ended up using? I have started with David Lu's arm_kinematics (http://www.ros.org/wiki/arm_kinematics), but there is also the OpenRave (http://www.ros.org/wiki/openrave)route.

--patrick

lnxfergy
04-06-2011, 12:52 PM
Patrick,

I'm using the arm_kinematics package -- it was just too easy to setup to skip over (and I'm not terribly worried about the speed). There's also the issue that Maxwell's arm is 5DOF, which means the openrave stuff probably won't work (requires 6DOF I believe, otherwise you're overconstrained).

Here's a brief excerpt from the Robogames Symposium paper on Maxwell:

We use the existing ROS arm_kinematics package to compute inverse kinematics (IK) solutions. However, as our manipulator has only 5 degrees of freedom, we cannot reach all 6-DOF poses. The X, Y, Z coordinates of the piece are therefore converted into a 6-DOF pose by several heuristics:


The yaw angle of the grasper is determined uniquely from X and Y coordinates of the grasper.
The roll angle of the grasper can be specified to maximize the grasp ability.
The optimal pitch angle of the grasper is an overhead grasp, however we can search the IK configuration space within a small region around overhead grasping to find a suitable solution.

-Fergs

Pi Robot
04-06-2011, 01:10 PM
Thanks Fergs--that's great to hear since I've already got some of the arm_kinematics package working. Pi also only has 5-DOF arms but also I'm not trying to grip anything yet--just place the hand at a target location. So I'll keep plugging away...

--patrick

Pi Robot
04-07-2011, 09:38 AM
Patrick,

I'm using the arm_kinematics package -- it was just too easy to setup to skip over (and I'm not terribly worried about the speed). There's also the issue that Maxwell's arm is 5DOF, which means the openrave stuff probably won't work (requires 6DOF I believe, otherwise you're overconstrained).
-Fergs

Fergs,

I'm running into an inconsistency with the arm_kinematics package that I can't figure out. I'm just trying to get started with the simplest test I can think of: first I run FK on one arm with all the joint angles set to 0 and I write down the resulting pose of the hand link using the shoulder link as the root. Then I take these pose values and feed it to the IK solver with the same root link and target link. I seed the solver with value 0 for all the joints which is actually the right solution. However, the solution returned is way off as shown below:


solution:
joint_state:
header:
seq: 0
stamp:
secs: 0
nsecs: 0
frame_id: ''
name: ['left_shoulder_pan_joint', 'left_shoulder_lift_joint', 'left_arm_roll_joint', 'left_elbow_joint', 'left_wrist_joint']
position: [0.63883168446830108, -0.0010338866667795534, -0.00020294896350323937, -1.0066250801356165, -0.3677934294831165]
velocity: []
effort: []
multi_dof_joint_state:
stamp:
secs: 0
nsecs: 0
joint_names: []
frame_ids: []
child_frame_ids: []
poses: []
error_code:
val: 1
Note that for some reason frame_id is null even though I have set it as follows:


request = GetPositionIKRequest()
request.timeout = rospy.Duration(5.0)
request.ik_request.pose_stamped.header.frame_id = 'left_shoulder_link'

Also, I believe the error_code val of 1 means a solution actually could not be found. Any initial thoughts on what I might be doing wrong. (I can of course provide more of the code I am using if needed).

--patrick

Pi Robot
04-07-2011, 10:06 AM
OK, ignore my previous question. Turns out, because I was using the extreme extended position of the arm for my test case, I had to copy and paste the FK pose with all the values after the decimals (I was using a 4-decimal approximation). If I then pull the target position of the hand somewhat inside of the extreme position, I get IK solutions nicely!

--patrick

lnxfergy
04-07-2011, 11:57 AM
I just posted this to my blog, and figured I would copy it here:

Next week is National Robotics Week (http://www.nationalroboticsweek.org/) and is going to be quite a busy week for me. Maxwell will actually be touring the country for NRW this year, making appearances at the following events:


Trinity Fire Fighting Home Robot Contest (http://www.trincoll.edu/events/robot/)- Saturday April 9 and Sunday April 10th.
Stanford Robot Block Party (http://cyberlaw.stanford.edu/node/6611) - Thursday April 14th
Robogames (http://robogames.net/) - Friday April 15th to Sunday April 17th. Maxwell is entered in the Best in Show category, as well as being featured in the Symposium.

Additionally, I'll be presenting a paper on a separate topic at the IEEE TePRA conference on Monday and Tuesday the 11th/12th, while Maxwell is in transit across the country.

-Fergs

Pi Robot
04-07-2011, 12:06 PM
Good luck Fergs. I hope to see you and Maxwell at RG on the Sunday and perhaps at the Stanford RBP if I can get off work for the afternoon.

--patrick

Pi Robot
04-27-2011, 09:19 AM
Lastly, there was a request for a parts list -- it's been up on the wiki for some time now: parts list (https://spreadsheets.google.com/pub?key=0AoEGbFgI5kOhdG9fRkFTVnc5VXkyZ0JYcW5laDE3N kE&hl=en&output=html). Unfortunately, it's already a bit out of date. Should be updated shortly.

-Fergs

Hey Fergs,

One part I could not find on your list is the step-up/step-down transformer you used to get 12V to the Kinect. Do you have a link to that part? Also, what are the pros/cons of that particular model?

Thanks!
patrick

Pi Robot
04-27-2011, 11:17 AM
Fergs,

I had an idea for Maxwell and Pi Robot for getting the arms and/or head up and down to different heights. How about using a "Johnny 5" type hinge mechanism integrated into your aluminum riser? Similar to this:

http://www.lynxmotion.com/images/html/build121.htm (http://www.lynxmotion.com/images/html/build121.htm)

only using Dynamixel servos and brackets instead? Given the right Dynamixels, there shouldn't be any problem lifting the weight, and the range of motion would of course be determined by the lever arms inbetween servos and riser. If you think this is feasible, it seemed to me it might be easier to construct than some kind of vertical pully/linear actuator/worm gear. And it would be potentially faster too.

Whaddya think?

--patrick

lnxfergy
04-27-2011, 12:31 PM
Fergs,

I had an idea for Maxwell and Pi Robot for getting the arms and/or head up and down to different heights. How about using a "Johnny 5" type hinge mechanism integrated into your aluminum riser? Similar to this:

http://www.lynxmotion.com/images/html/build121.htm (http://www.lynxmotion.com/images/html/build121.htm)

only using Dynamixel servos and brackets instead? Given the right Dynamixels, there shouldn't be any problem lifting the weight, and the range of motion would of course be determined by the lever arms inbetween servos and riser. If you think this is feasible, it seemed to me it might be easier to construct than some kind of vertical pully/linear actuator/worm gear. And it would be potentially faster too.

Whaddya think?

--patrick

Maybe.. I've pondered this before (My favorite (only?) example of this is SmartPal V (http://www.engadget.com/2007/12/03/smartpal-v-robot-now-with-additional-lumbar-units/)). It would certainly be faster, however there are a few issues to contend with:


You're going to need a really big servo. We're looking at a lever arm in the 18-36" long range, and quite a bit of weight out at the end of that lever. My bigger concern here is that these servos do get hot when under a lot of load, and especially in the case of the larger RX and EX servos, tend to not dissipate heat until they are powered off.
You're going to need a lot of ballast on the back of the base -- to offset the big change in COG when bending over.

-Fergs

Pi Robot
04-27-2011, 06:52 PM
Yeah, I see what you mean. The Johnny 5 probably doesn't weigh much above the waist and the lever arm is fairly short...

--patrick

lnxfergy
04-28-2011, 11:02 PM
I created a short demo video of Maxwell using the ROS navigation stack and pocketsphinx (the "goto" demo) for a talk earlier today, figured I would share:


http://www.youtube.com/watch?v=8aABif3W6MY

-Fergs

Peter_heim
04-29-2011, 07:21 AM
Hi Fergs
very impressive video i have tried goto.py and it works for me. In the video you had a mike on ( that was you?) was that for the video or to control maxwell?
I have been playing with voice recognition and it works when within say 3 feet but further than that and back ground noise i have found it to be a pain, I seen some packages (hark) that localize sound have you any thoughts how to improve hearing (for want of a better word). I haven't done any voice training yet i want it to work with a range of people ie my 9 year old son and of course the wife.

Peter

Pi Robot
04-29-2011, 08:28 AM
I created a short demo video of Maxwell using the ROS navigation stack and pocketsphinx (the "goto" demo) for a talk earlier today, figured I would share:

YouTube - Maxwell Navigating (http://www.youtube.com/watch?v=8aABif3W6MY)

-Fergs

Sweet! Inspired by your video, I now have my Bluetooth headset working with pocketsphinx. The only problem is that when I say "Please make the bed", Pi Robot just sits there and stares at me. ;)

--patrick

lnxfergy
04-29-2011, 08:56 AM
Hi Fergs
very impressive video i have tried goto.py and it works for me. In the video you had a mike on ( that was you?) was that for the video or to control maxwell?
I have been playing with voice recognition and it works when within say 3 feet but further than that and back ground noise i have found it to be a pain, I seen some packages (hark) that localize sound have you any thoughts how to improve hearing (for want of a better word). I haven't done any voice training yet i want it to work with a range of people ie my 9 year old son and of course the wife.

Peter

Yes, that was me, and I was wearing a bluetooth headset (this one). We have been working on using a shotgun microphone mounted on the robot, which introduces new issues, which we are working on solutions to.

Training pocketsphinx is not an easy undertaking,

-Fergs

lnxfergy
04-29-2011, 08:58 AM
Sweet! Inspired by your video, I now have my Bluetooth headset working with pocketsphinx. The only problem is that when I say "Please make the bed", Pi Robot just sits there and stares at me. ;)

--patrick

Yeah, Max does the same. The next video will show him handling things such as "go to the lab and get me X"...

-Fergs

Pi Robot
04-29-2011, 05:55 PM
Hey Fergs,

Where is the "goto.py" pocketsphinx script you guys are referring to? I have the latest pocketsphinx from http://albany-ros-pkg.googlecode.com/svn/trunk/rharmony/pocketsphinx and it does not appear to contain a script by that name. (Though the voice_cmd_vel.py script works nicely.)

Thanks!
patrick

lnxfergy
04-29-2011, 05:59 PM
Hey Fergs,

Where is the "goto.py" pocketsphinx script you guys are referring to? I have the latest pocketsphinx from http://albany-ros-pkg.googlecode.com/svn/trunk/rharmony/pocketsphinx and it does not appear to contain a script by that name. (Though the voice_cmd_vel.py script works nicely.)

Thanks!
patrick

Goto is in Maxwell's navigation package.

-Fergs

lnxfergy
05-06-2011, 04:28 PM
Just a note, Maxwell's software has moved into the vanadium-ros-pkg repository. The fergy-ros-pkg repository will disappear in a few days once the indexers have caught up.

-Fergs

SuperMiguel
12-28-2012, 08:36 PM
I thinking about a maxwell too =) similar but different =) just have few questions...

What are you using to control the wheel motors? the arbotix built in motor controller??