Slashdot Mirror

Yes, you can learn a lot of robotics without actual hardware. I develop software for self-driving vehicles, and spend 95% of my time away from the hardware!

ROS + Gazebo will let you assemble a robotics software stack and explore different planning and control algorithms in simulation: http://gazebosim.org/ and http://www.ros.org/

If you want to explore perception and computer vision, take a look at OpenCV ( http://opencv.org/ ) and the tutorials there. The great thing about computer vision is you can run your software against the standard research sets or images you pull off Flickr.

Point Cloud Library is a nice package for looking at 3D laser data (but has some numerical quirks): http://pointclouds.org/

I would definitely take a look at some MOOCs, Andrew Ng's Machine Learning at Coursea (https://www.coursera.org/learn/machine-learning) or the MIT Courseware ( http://ocw.mit.edu/courses/ele... )

>Why sould a GPU manufacturer spend a lot of time supporting such a small user base

I don't know, maybe because most super computers on the fucking planet use GPUs? Why would scientists want a GPU manufacturer to support the operating system they do most of their work on? Oh, I can't think of a reason.

Meanwhile, we're trying to do some work in ROS. I certainly don't want CUDA cores to help speed up the processing and filtering of tens of thousands of LIDAR points. Nor could I possibly use shaders for anything outside of gaming.

This much sarcasm is killing me. Please get better opinions before I die.

In their introduction they use the term meta-operating system.

ROS is an open-source, meta-operating system for your robot. It provides the services you would expect from an operating system, including hardware abstraction, low-level device control, implementation of commonly-used functionality, message-passing between processes, and package management. It also provides tools and libraries for obtaining, building, writing, and running code across multiple computers. http://wiki.ros.org/ROS/Introd...

Are you familiar with ROS, http://www.ros.org/ ? It's basically a set of libraries for various robotics tasks and sounds like what you're describing.

For those like me who didn't know, it is the set of Linux packages enumerated here.

The OSU Open Source Lab is picking up the Robot Operating System's web presence (ROS being the software that powers WillowGarage's units). It's been a long process, as they have a lot of moving pieces that we're integrating, but hopefully the entire setup will be completed next week and we'll have an announcement to make.

A "Robotics" project sounds way to generic, A little more detail on your end goal would help you focus better. If you want a premade solutions with all interfaces I'd start here Lego Mind storms If you want to try your hand at control algorithms without spending a penny I'd start here (sharp learning curve) http://gazebosim.org/wiki/DRC/Install If you want to visually do something with your robot i'd start here, various boards and controls are included. http://www.roborealm.com/ If you want a bit more advanced hardware I'd start here http://www.ros.org/wiki/Robots For pure visual processing fun, this actually is rolled into ROS and DRC sim i believe http://opencv.willowgarage.com/wiki/

From least to most expensive, here are some mid budget items that would get me excited:
In the $200 range, and because I enjoy programming:

• - A kinect with either the Microsoft API
• - A primesense 3D sensor with the OpenNI API

For super extra bonus fun, integrate the sensor with the robot operating system so that you can use their cool 3D visualization tools.

Another sweet toy, slightly more expensive at $400, is the Lytro camera.

My final toy choice would be a very open 3D printing platform, like the 3rd generation Solidoodle, which at $799 is actually only accepting pre-orders now.

If I had all that stuff, and had enough time to play with it, I'd be pretty happy.

From least to most expensive, here are some mid budget items that would get me excited:
In the $200 range, and because I enjoy programming:

• - A kinect with either the Microsoft API
• - A primesense 3D sensor with the OpenNI API

For super extra bonus fun, integrate the sensor with the robot operating system so that you can use their cool 3D visualization tools.

Another sweet toy, slightly more expensive at $400, is the Lytro camera.

My final toy choice would be a very open 3D printing platform, like the 3rd generation Solidoodle, which at $799 is actually only accepting pre-orders now.

If I had all that stuff, and had enough time to play with it, I'd be pretty happy.

The nature of my work requires it. I'm a robotics engineer and I do work from the embedded level all the way up to the high level behaviors level. For the latter work, we use the ROS framework, which relies heavily upon various Linux libraries. We also use Android tablets and phones as remote interfaces for some of our robots, so I have experience developing for them, interfacing with the cameras and inertial sensors mostly.

Every single time there's a Kinect story, someone has to ask this question. My cynical side wonders whether they'd still be doing this if the Kinect wasn't a Microsoft product, but here's a comment I posted a while back that explains why, even though it may be overrhyped as a video game peripheral, the Kinect is damned cool for robotics applications. The robotics community has, almost overnight, dropped work on many other kinds of input sensors to focus on the Kinect because it's so much more useful (see openni_kinect on ROS for the primary driver, although we'll see if this continues to see love now that an official SDK is out).

So by that definition, a Yugo is revolutionary.

Yugo didn't decrease the cost of an automobile 10 fold while simultaneously increasing the quality. It isn't JUST that the Kinect is so cheap. I can buy an sharp IR range finder for $15, 10 times cheaper than even the Kinect. But in terms of resolution, it's not 24 times better, like the Kinect is compared to the IFM sensors I'm currently using. If Yugo cost $300 and got 100mpg, then we'd be talking about revolutionary.

A perfect example of this happening before is sitting right in front of you. Computers used to take up entire rooms. These had niche application in business, government, and academia, but the computer revolution didn't happen until they were cheap enough for anyone to afford. Not only that but as they became cheaper they also became better!

In terms of your definition, Kinect meets this perfectly. It has had a major sudden aspect on an aspect of human endeavor, specifically robotics. Society has changed in that anyone can now afford to integrate what used to be expensive data into their robotics project. It's even further revolutionizing academic research, which you claim has been doing this all along, and therefore should be no big deal. We now all have the same sensors. I can develop an algorithm in my lab, publish it on the internet, and anyone in another research lab can download it and use it without worrying about configuration, compatibility, differences between sensors. Not only that, any hobbyist can download my state of the art algorithm and use it on their personal robotics project. This isn't something that might happen one day, it's .

I am not doubting that the Kinect is changing the game. I just question whether it is revolutionary or not.

Kinect is changing the game but it's not revolutionary? What's your definition of a revolution then? Before the kinect, it cost me close to $10,000 for a good 3d point cloud data. If I had more room on my robot, I might put a Hokuyo LIDAR on a pivot but that still put me back 6 grand. Today I use industrial sensors from IFM, re-purposed for Robotics. They cost about $1500, and only provide 50x64 pixels of range data, as compared to the Kinect's 320x240.

So the cheapest feasible sensor I can buy costs $1500. So here comes Microsoft. They're selling a sensors 10 times cheaper with 24 time the resolution. Now any old schmuck can buy this and test their idea for a new image segmentation algorithm. This has NEVER been possible before.

So yeah, Kinect is changing the game. That's the definition of a revolution. Just because it was done in a lab before by Ph.D.s after 10s of thousands of dollars of time, effort, and equipment doesn't diminish it. If a company started selling robot cars to the public, that would be revolutionary too, even though we can do that in the lab (for $1,000,000+).

And Microsoft can't get all the credit; none of this would be possible without ROS and the amazing Point Cloud Library. This is a second component of the kinect revolution, which, in itself is revolutionary.

I am not doubting that the Kinect is changing the game. I just question whether it is revolutionary or not.

Kinect is changing the game but it's not revolutionary? What's your definition of a revolution then? Before the kinect, it cost me close to $10,000 for a good 3d point cloud data. If I had more room on my robot, I might put a Hokuyo LIDAR on a pivot but that still put me back 6 grand. Today I use industrial sensors from IFM, re-purposed for Robotics. They cost about $1500, and only provide 50x64 pixels of range data, as compared to the Kinect's 320x240.

So the cheapest feasible sensor I can buy costs $1500. So here comes Microsoft. They're selling a sensors 10 times cheaper with 24 time the resolution. Now any old schmuck can buy this and test their idea for a new image segmentation algorithm. This has NEVER been possible before.

So yeah, Kinect is changing the game. That's the definition of a revolution. Just because it was done in a lab before by Ph.D.s after 10s of thousands of dollars of time, effort, and equipment doesn't diminish it. If a company started selling robot cars to the public, that would be revolutionary too, even though we can do that in the lab (for $1,000,000+).

And Microsoft can't get all the credit; none of this would be possible without ROS and the amazing Point Cloud Library. This is a second component of the kinect revolution, which, in itself is revolutionary.

Do you have any idea what the kinect is? Eyetoy is an overpriced glorified webcam with some object segmentation and gesture recognition behind it.

Kinect is a stereoscopic camera combined with an infrared depth sensor. This provides the xbox with a 640x480 RGBD image it uses for full-body skeleton tracking, facial recognition, gesture recognition.

If that wasn't enough, it contains motors, an accelerometer, and a microphone array for body tracking and voice recognition.

With very little modification, the same sensor used for playing silly dance games on the xbox can be used for everything from teleoperating humanoid robots to an augmented reality midi device.

Seriously, comparing the kinect to an eyetoy is just so far off base it's not funny.

Do you have any idea what the kinect is? Eyetoy is an overpriced glorified webcam with some object segmentation and gesture recognition behind it.

Kinect is a stereoscopic camera combined with an infrared depth sensor. This provides the xbox with a 640x480 RGBD image it uses for full-body skeleton tracking, facial recognition, gesture recognition.

If that wasn't enough, it contains motors, an accelerometer, and a microphone array for body tracking and voice recognition.

With very little modification, the same sensor used for playing silly dance games on the xbox can be used for everything from teleoperating humanoid robots to an augmented reality midi device.

Seriously, comparing the kinect to an eyetoy is just so far off base it's not funny.

It takes maybe 4 commands to get this working.

sudo sh -c 'echo "deb http://code.ros.org/packages/ros/ubuntu maverick main" > /etc/apt/sources.list.d/ros-latest.list'
apt-get install ros-cturtle-ros
apt-get install ros-cturtle-kinect

That will get the Kinect working. Then you just need a GPS and the appropriate transform between the GPS antenna and the sensor (easiest to put the antenna on top of the kinect), and you're done.

And I don't see what me being German has anything to do with this.

This is trivially done with ROS and the Kinect stack. I went out and bought a kinect and plugged them into my robot platforms, provided a transform between robot base and kinect sensor, and was done. It's a great application and anyone who owns a kinect should try it out, but at this point it's trivial and hardly worthy of a /. post.

While the work the MIT student did is noteworthy, it's really quite trivial thanks to ROS. I do robotics research using ROS, and SLAM, navigation, planning, etc. are all handled by ROS automatically as long as you provide the appropriate data streams. It's really as simple as plugging in a device. Even the gesture recognition is handled by the kinect driver and issuing commands from gestures is trivial at that point.

I think the real recognition should be given to the group at CCNY (no I don't got school there) who did the work of getting the kinect driver working in ROS in the first place, and aren't even mentioned in this article.

http://www.ros.org/wiki/kinect_node

With the calibration the accuracy of Kinect is much improved.. and ROS has algorithms that can do this automatically for anyone lucky enough to have a manipulator - speaking of which, when is Microsoft coming out with a $150 robotic arm? :)

I admit it I didn't RTFA... but it sounds like the Willow Garage stuff... http://www.ros.org/wiki/ ?

(http://www.willowgarage.com/ for the record is their main site).