We have a Linux Version and a Windows Version. The Windows version is really so that people can use the Windows gui and 802.11 drivers for wireless control. The real motor control is done in a separate hardware box that we provide. The real breakthrough with this product is not in the robot construction so much but more in the software, in particular the vision algorithm that lets the robot navigate autonomously and recognize objects and people.
Yes, it does have an optional arm, and it can do amazing things. Yes, we have a pretty flexible spec. We also have a simpler windows version that allows you, yes, even you!, to program things very easily. Did you see the picture of the interface at our web site?
http://er1.evolution.com/control/
Let me know what you think. And if it's not completely obvious how it works and what it does then we blew it!
The robot has 4" wheels, so it can only go over obstacles abotu 3/4" to 1" tall. It can most likely make it over a 3/4" threshold. It is really good at navigating obstacles. It's pretty good about dirt (it uses scooter rubber wheels). It cannot climb stairs. It's cheap enough to buy one for each floor! It can ferry parts around really well because of the vision-based navigating.
I think the "bring me a beer" is not that useful, although neat, but to use that same functionality to do something more valuable, like bring me some medicine (the robot can read medicine labels, so this might be very useful for elderly) or bring me that package (for inventory) or bring this fedex to room 207 might be much more useful. We were just showing the beer example to show how the robot could navigate based on vision and how it could recognize similar looking objects even in random lighting and settings (of the trade show booth).
The reason that the processor requirement is so high is because the vision algorithm is doing massive stuff on each frame that comes in from the camera. This kind of robust vision has never been done before on ANY type of machine before, so to pull it off on an under 1ghz machine -- I think you will find it very impressive when you see it.
Yes, it requires a pretty hefty Pentium to do the real time vision, at least 800 mhz or so. We are doing processing at 5 fps of up to 10,000 objects, using up to 1,000 features of each object, so it's very cpu intensive, even after extensive mmx optimization. But it works very well, and it can recognize extremely accurately even in varying lighting conditions and at offset angles. We have a rotation invariant, scale invariant, affine transform invariant, and ligthing invariant transform that we do to the features in the image as we process them and that allows us to do the check so quickly and with accuracy in a real world situation.
Yes, the ER1 Robot also has voice recognition. It's pretty cool. You CAN say "play U2" or "play Hey Jude" and it begins playing from an mp3 collection virtually instantly. You can branch off of voice or images/vision. Images can be of objects, books, or people.
The new Evolution Robot ER1 has an optional arm, so you can pick up things or actuate things in the environment. At E3 where we unveilded this new robot, we are doing a demonstration of the famous "get me a beer" application. In it, we show the robot an empty beer can or bottle, it recognizes it, then goes to the fridge, using vision, and then picks up the correct beer, and brings it back to you. I wish there were a way I could post an image or a movie of it doing this, it's quite impressive.
Slashdot Mirror
We have a Linux Version and a Windows Version. The Windows version is really so that people can use the Windows gui and 802.11 drivers for wireless control. The real motor control is done in a separate hardware box that we provide. The real breakthrough with this product is not in the robot construction so much but more in the software, in particular the vision algorithm that lets the robot navigate autonomously and recognize objects and people.
Yes, it does have an optional arm, and it can do amazing things. Yes, we have a pretty flexible spec. We also have a simpler windows version that allows you, yes, even you!, to program things very easily. Did you see the picture of the interface at our web site?
http://er1.evolution.com/control/
Let me know what you think. And if it's not completely obvious how it works and what it does then we blew it!
The robot has 4" wheels, so it can only go over obstacles abotu 3/4" to 1" tall. It can most likely make it over a 3/4" threshold. It is really good at navigating obstacles. It's pretty good about dirt (it uses scooter rubber wheels). It cannot climb stairs. It's cheap enough to buy one for each floor! It can ferry parts around really well because of the vision-based navigating.
I think the "bring me a beer" is not that useful, although neat, but to use that same functionality to do something more valuable, like bring me some medicine (the robot can read medicine labels, so this might be very useful for elderly) or bring me that package (for inventory) or bring this fedex to room 207 might be much more useful. We were just showing the beer example to show how the robot could navigate based on vision and how it could recognize similar looking objects even in random lighting and settings (of the trade show booth).
The reason that the processor requirement is so high is because the vision algorithm is doing massive stuff on each frame that comes in from the camera. This kind of robust vision has never been done before on ANY type of machine before, so to pull it off on an under 1ghz machine -- I think you will find it very impressive when you see it.
Yes, it requires a pretty hefty Pentium to do the real time vision, at least 800 mhz or so. We are doing processing at 5 fps of up to 10,000 objects, using up to 1,000 features of each object, so it's very cpu intensive, even after extensive mmx optimization. But it works very well, and it can recognize extremely accurately even in varying lighting conditions and at offset angles. We have a rotation invariant, scale invariant, affine transform invariant, and ligthing invariant transform that we do to the features in the image as we process them and that allows us to do the check so quickly and with accuracy in a real world situation.
Yes, the ER1 Robot also has voice recognition. It's pretty cool. You CAN say "play U2" or "play Hey Jude" and it begins playing from an mp3 collection virtually instantly. You can branch off of voice or images/vision. Images can be of objects, books, or people.
The new Evolution Robot ER1 has an optional arm, so you can pick up things or actuate things in the environment. At E3 where we unveilded this new robot, we are doing a demonstration of the famous "get me a beer" application. In it, we show the robot an empty beer can or bottle, it recognizes it, then goes to the fridge, using vision, and then picks up the correct beer, and brings it back to you. I wish there were a way I could post an image or a movie of it doing this, it's quite impressive.