Rat brain controls mobile robot

[R2-D2]

let's implement a cheese following algorithm!

[nano]

This is really interesting. And I find it amazing that they are "growing" neural control circuits for these robots.

http://blog.wired.com/defense/2008/08/r ... t-vid.html

However, the scientists do still need to figure out were to "tap into the system" to get the desired robot reaction (obstacle avoidance). This gives me the feeling that there is no learning or adaptation but just an opportunistic use of the electric signals from these neurons to create a sort of braitenberg vehicle by tapping in at the right spot. I guess the next step would be to figure our how the robot/rat can have a sense of what its "purpose" is (avoid obstacles) so that it can autonomously generate an effective neural pathway to its actuators and sensors.

Is there a publication yet? I noticed that this work was done by Prof. Kevin Warwick, the Cyborg man in the UK, but couldn't find any ref on the web.

Final question, is this still a robot or is it an augmented rat?

[epokh]

The mediatic flow is exagerated.
The robot control experiment is rubbish, I will write my critic tomorrow, now time to sleep.
Here we are:
1) how can you track a 100,000 neuron network using only 8x8=64 electrodes ?
The microarray matrix is well know and used in previous experiments where you can grow neurons in a bath.
2) what's the percentage of excitatory and inhibitory neurons that are being transplanted from the fetus?
It seems they are all excitatory: so basically when sensors are activated the robot is randomly driving back.
The baseline activity should be responsible for the forward motion, so how are the output neurons connected to the motor output?
I bet my balls is crafted.
3) how can you learn a behaviour having only 1 type of sensor (long range) and of course no reward signal?
I would not have been suspicious if they were using 2 type of sensors so that the network can self-organize using the temporal relation between them
4) the fetus develop it's neural network structure in "chilled" environment, using their approach of totally developing plastic connection in a very complex environment like the avoidance scenario is very bad.
It's true children can sense something from outside the amniotic bag but is very mild.
5) there will be no paper I guess, all the methods are well know, if they manage to explain one of the previous questions maybe they will get a publication.

It's so bad seeing all this media focusing on him, in our lab we used brain stimulation to decrease schizophrenia in rats, and nobody heard about it. bah

[nano]

Interesting critics epokh, makes me think... Sadly, I'm no expert and the media coverage is quite confusing and imprecise.

I just found an article from the Potter Group at Georgia Tech on the subject:
http://www.ploscompbiol.org/article/inf ... bi.1000042

In the end, while the behavior of the robot is "crafted", maybe the whole point is in understanding how "crafting" a rat brain can be done in the first place?

[epokh]

Sorry for the late reply (I was goddam busy with some other meetings).
What you have cited is a serius article and as you can read:

The closed-loop design we discuss here consists of four unique elements:

1.Patterned stimulation to induce network plasticity. This low-frequency (~3 Hz) training stimulation differs from most studies of cultured networks, where plasticity was induced by high frequency tetanic stimulations [15],[17].
2.Continuous low-frequency background stimulation (~3 Hz) to stabilize accumulated plasticity [19], which is analogous to continuous sensory inputs and ongoing processing in the brain.
3.
Population coding for motor mapping. Population coding is considered a robust means to represent movement directions in the primary motor cortex [21].
4.
Adaptive selection of training stimulation. Because the connectivity in a cultured network is not predictable, the effects of a given training stimulation cannot be known a priori. Thus we delivered training stimulation contingent on the animat's performance in order to direct changes in network connectivity that further shift the animat's behavior toward the desired behavior.

As you can read in point 1 and 4 they are training it by repeating a typical pattern (in this case an attraction behaviour) to induce LTP.
You can't put a system like that in the environment suppose it will learn a behaviour (in worwick case avoidance) without teaching it.
And just to say that i'm right I cite:

We developed tools [1]–[3] for closing the sensory-motor loop between a cultured network and a robot or an artificial animal (animat) [4] in order to study learning directly through behavior of the artificial body and its interaction with its environment

So this technique of closing the loop of a cultured neural network is NOT novel at all:
[4]Meyer JA, Wilson SW (1991) From Animals to animats: Proceedings of the First International Conference on Simulation of Adaptive Behavior. Cambridge: MIT Press.

[Markus Waibel]

Interesting discussion.

The robot is indeed all over the web - and was just featured in this Robots week's news. Looking forward to some more opinions!

Here are some more links for background infos:
Roland PiquePaille: http://blogs.zdnet.com/emergingtech/?p=1009
New Scientist: http://technology.newscientist.com/chan ... obots.html
Engadget: http://www.engadget.com/2008/08/13/uk-r ... cal-brain/

[epokh]

The newscientist article is rather strange, what I know from working memory and cortical circuits is that:
a)
a "bursting" network doesn't store any information at all: synchronization it means your memory is gone
b) there should be a baseline of random spiking activity: white noise like when inputs are absent
c) when inputs are present there should be some pattern propagating into the network
d) I'm curious to see STDP like changes on the synaptic connections after pattern stimulation/ learning

The main problem of this circuit is that the dopamine signal is absent is so damn hard to get a stable framed learning:
basically the network is basically learning online continuously therefore the training patterns must be provided carefully (like the guy says injecting pulses of currents as signal patterns).

Oky those are comments for the general public, so I'm waiting for the real journal article.

[hthth]

To pitch some similar projects into the discussion: There was a research project I remember reading about a few years back that involved training rat neurons to fly a fighter jet (in a simulation). If I remember correctly, the neurons were reportedly trained to fly even in tough weather conditions.

Here we go, they used 25,000 neurons and 60 electrodes.

Then there was the robot controlled by a moth brain earlier this year.

Has anyone come across articles comparing the methods of these projects with the new one?

[JSM4]

im no genius nor do create robots.. but in my fair understanding, this creation of robots, with actual animal brain is just so fascinating! it really blows my mind, in all sorts of levels.. i salute you guys forconducting such great inventions and innovations!

my boy is in MIT, and i hope he's doing justice to his studies, and become just like you guys!
great job!