Custom firmware for the original AR.Drone.

Low-level Linux exploration of an original AR.Drone.

Article announcing the MK802 AllWinner ARM USB stick computer. Currently selling for $74.

Relatively cheap camera module for the Gumstix platform.

"The fish use micro-electrode arrays to sense contaminants. In their current form they can detect phenols and heavy metals such as copper and lead, as well as monitor oxygen levels and salinity.

But the team has tried to build in flexibility.

Dr Speller explains: "We have designed it so you can pull out the chemical sensor unit, and put in different ones for something else, such as sulphates or phosphates, depending on the environment that you are monitoring," he explains.

One they've sniffed out a problem, the fish use artificial intelligence to hunt down the source of pollution.

They can work alone or in a team, communicating with each other using acoustic signals and they can continuously report back to the port.

"In the future, what I'd also like to see is not just a single task robot, but robots that can multitask - robots that can do search and rescue, monitoring for underwaters divers, at the same time as tracking pollution."

[object Object]

this is AMAZING.

"If these researchers were able to look at it this way -- as the control of perception -- I think there would be a couple of things they could try that might improve things. One place to look for possible improvements is in the display of what the robot is doing: perhaps add lines and arrows of varying length and size indicating the states of the perceptual aspects of the robot's behavior that are influenced by the variations in neural firing rates."

Circular BEAM "walker"

a 3D occupancy grid mapping approach, providing data structures and mapping algorithms in C++. The map implementation is based on an octree

"If I find the time, I might try to post some tutorials on using Ceres. Because I believe this is one of the most powerful tools in modern engineering, and no one ever taught it to me in undergrad or high school. It's like the difference between doing long division by hand and then being handed a calculator." Wow.

"1) 2-dimensional decoders, where an individual controls a mouse with their mind, have existed for a while. What makes this really cool is that there are actually quite a few degrees of freedom in a robot arm, but they're basically using the same hardware. So the influence of software/algorithms in this case is pretty fundamental. There are a lot of papers on improved neural-decoding methods that at first glance appear really dry, boring, not the `sexy' kind of science with huge breakthroughs, but they end up being crucial to good performance as the complexity of the robot grows.

2) One participant was implanted with the electrode array 5 years before the study, and had the injury 10 years before that. Usually the signals don't last that long in monkey models. And we know that your cortex changes with disuse, so it's awesome that they were able to get usable signals so many years later.

What needs to get better are a couple things:

First, these decoders aren't perfect yet. What the Wired article didn't tell you is that the performance for the woman's implant was around 20-50% successful trials (still awesome from a comparison to no interaction)[1].

Second, incorporating sensory feedback is another challenge that is really hard to address, but also very important. Imagine building a robotic controller in which the only information you received about the robot's position was visual. That's the way this works. If we find a reasonable way to mimic sensors of muscle extension (a proxy for joint angle) then we can create more controllable devices."