[SoftwarePatent]

Is there a sense in which software is what is holding back robots? We have all watched pretty cool robots in videos rolling around, walk on two legs, etc. What we need is awesome AI to put in the existing hardware. Any experts want to chime in?

[mdda]

I don't claim to be an expert, but from what I've seen there's a typically a huge gulf between the software and hardware.

Until recently, it's been difficult to put a lot of CPU on a mobile robot, so the software developer really has to be a programmer with an embedded systems mindset. For instance, the Arduino is a relatively recent phenomenon, and that's seriously under-powered if you're interested in high-level planning, or vision. But the GumStix or BeagleBoard look interesting, since there's a pretty accessible tool-chain available.

On the hobbyist front, most of the people setting out to build (for instance) humanoid robot kits have a lot of fun with getting the hardware together, but realize that dabbling in software really doesn't get you very far.

For the software-mindset hobbyists (like me), it's a lot more practical to do robot experiments virtually - since as soon as one gets out the soldering iron, the whole weekend disappears instantly.

"Hardware is tough", and (unlike software) iterating the design/plan is extremely time-consuming. OTOH, once a few decent platforms appear (in the $1000 ballpark), I guess software people will swarm in and push things into exponential mode, rather than what seems like linear mode right now.

[baalexander]

I recently bought a TurtleBot (turtlebot.com), a ~$1,000 robot with a netbook, kinect, gyro, and roomba. The TurtleBot platform combined with the Robot Operating System (ros.org) - a popular, open-source robot framework - is an affordable way to test out and learn many cutting edge robotic algorithms like room mapping and navigation. It even supports telepresence.

I consider myself software oriented too, and recommend it to others who are looking at going beyond a basic microcontroller based robot, but without all the soldering.

[nknight]

> Until recently, it's been difficult to put a lot of CPU on a mobile robot

Define "mobile"?

Household and other single-site robotics applications should not be severely constrained by the advancement of embedded computational hardware. Wireless transceivers can provide the bridge between the limited embedded systems in the physical-actor hardware in the living space, and the heavy computational hardware in a big box in the closet.