Tuesday, January 15, 2008

monkey-controlled robot destroys los angeles

This story was printed today in the New York Times science section: "Monkey's thoughts propel robot, a step that may help humans." I'm always on the lookout for anything neuroscience-related that seems to fuck with our notion of what's science fact and what's science fiction, and this one certainly fits the bill. As I understand it, the scientists implanted some recording electrodes into the motor cortex of a monkey, somehow translated this code of electrical activity into a set of instructions for a robot (?) such that when the monkey walked, the robot walked in a similar fashion. This, by itself, is pretty impressive. But the experimenters were also able to get the monkey to make the robot walk by using only its thoughts (and some clever operant conditioning), even when it wasn't walking on the treadmill. Here's a schematic that will help you make sense of what I just said (and visit the link for full details):

I am actually astounded by this. I really had no idea we, as a scientific community, had the knowledge necessary to make this possible. There are a couple of issues here that really stand out and beg for further information:

1. Where exactly in the motor cortex were these electrodes? It sounds like the team was trying to capture the neural code of not movement itself, but motor-planning. That's premotor cortex, rather than primary motor cortex. Premotor cortex is also where you find lots of mirror neurons, which fire when an animal either does something or watches another animal engage in the same behavior.

2. How the hell did they translate this neural code into a set of motor instructions for the robot? Quoting from the article:

"The video and brain cell activity were then combined and translated into a format that a computer could read. This format is able to predict with 90 percent accuracy all permutations of Idoya’s leg movements three to four seconds before the movement takes place."

Ummm, ok, but... what? FP/MC, got any thoughts here? Cracking the neural code in this fashion is perhaps the defining problem in computational neuroscience, and it seems like these guys have done it (at least with motion). Am I just naive to what's going on in my field? (probably)

If you're wondering why you'd even want to do something like this, well... I'm a little disappointed by your lack of imagination. Besides the eventuality described in the title of this post, you could use this technology to 1)
provide people who are paralyzed with robotic helpers, 2) control robotic movements in dangerous environments (e.g. mining delithium crystals on asteroids), 3) play a much cooler version of Team Fortress 2.


  1. Best I can tell, you'd do some sort of PCA on all them there neurons and find out what projected dimensions / 'factors' match up with the movement best. Then, its pretty straight forward to just tell the robot to use whatever motor program it has. This would be my quick and dirty guess.

    What I wonder is how direct the mapping is from the motor strip to the robot's leg controllers. I.e. - do they have just a simple motor 'program' that is a gait? If that's the case you can just say "on/off" and that'd be sort of boring. But if it's something more complicated then I'd be interested in the mapping- what sorts of 'inputs' does the robot have and how do those map to the 'outputs' of the monkeybrain2000.

    Pretty spookyass anyway.

  2. If Nicolelis did something similar to his 2000 Nature paper, then he used ensemble recordings from premotor and parietal cortex and applied both linear and neural network algorithms to predict position and velocity accurately and in real-time. It is pretty well explained in the paper and it is totally awesome. I am sending that paper to HHL if anyone wants it.

    And I like Flip's question a lot because you need to know 'position and velocity' a million times over to make bionic legs that walk and climb and kick. But that information is just enough to be punched by my new bionic club-hand.

  3. given the nature of the media source, it's obviously tough to make any real judgements regarding the methods used. that nothwithstanding, i had a similar reaction...

    in some ways this experiment does seem to have ostensibly 'cracked the neural code'. but i think the problem is that in this context (i.e., monkey to robot), the 'cracked code' doesn't have to remain 100% faithful...it's not being mapped onto another monkey brain. not that what they did was necessarily trivial, but i think what they would minimally have to do to pull off such a stunt is just find some reliable neural marker of 'move right leg, move left leg' in the monkey (could just be purely correlational neural firing, for instance), then send that signal in whatever form you need it to a receiver that will be responsible for executing the corresponding movements in the robot. i will say it's still impressive they appeared to find reliable neural markers of movement as well as pull off that kind of timing.

    i'm with you on the premotor v. motor cortex thing. seems like the mirror neurons that presumably would be responsible for the 'thought walking' would be in premotor cortex, not motor cortex. but given the reciprocal connective nature of the brain, wouldn't be out of the question that there could be some cells in motor cortex also involved in 'thought movement' but not real movement. but i think the times reporter probably fucked it up. look for an errata.

    to me, the real cracking of the neural code will come when we can take those signals from one brain, send them to a relay, and then send a meaningful representation to another brain so that we can deliver sight to the blind, feeling for the tactiley deprived, a heart for the tinman, etc..

    i will say this is certainly running up against my knowledge of how this shit could potentially work (actually works?), so i'm definitely no authority. maybe i'm missing something.

    either way, it would be bad ass to see some form of japanese apocalyptica go down on the streets of LA via a government designed 50-foot mutant chimp.

    by the way, it's total bullshit there has been no mention of 'DR' in any of your gaming related blogs. i don't even know you, man.

    deadbeat dad...that's messed up.