3 ms·

Designing a Collision Avoidance System by Watching Locusts Watch Star Wars

5 points by vezycash 8y ago

cimmanom 8y ago
That's kind of cool. But I hope we wouldn't be satisfied with a 91% collision avoidance rate.
tonyarkles 8y ago
Ooooh! Story time! In 2005 and 2006, I worked in a biology lab that worked on similar things as Dr. Rind's lab. I don't recall if we collaborated directly, but there was definitely a lot of overlap and I think we shared papers back and forth. When I arrived, the biologists had rigged up a pretty amazing setup. They had a system that was measuring locust wing-muscle activations, and feeding that in through a joystick port to Descent (yes, the game). This was a more closed-loop version of the "watching Star Wars" stuff talked about in the article. The system had a couple flaws, though. The biggest being that it was being projected onto a hemispherical dome that the locust sat in, but the scene wasn't perspective-corrected for the dome. It still worked, sort of though; you could watch the insect successfully navigate around obstacles, but it was pretty crude. Over the two summers (I was a student at the time), I did the majority of development on a more controlled and perspective-corrected system for them. It included a simple scripting language (since I had, you know, just finished my compilers course) for the scientists to describe specific scenes to project for the insect, instead of just using the pre-made Descent levels. It did some really cool projecting mapping, such that, given a known position of the insect's eyes relative to the hemispherical screen, we could project perspective-correct 3D imagery. It had a USB device for more predictable input than the joystick port. And it included time synchronization signals that could be fed back into the muscle-impulse-capture hardware to more easily correlate events in the 3D scene with muscle events. I haven't kept in touch with the lab much, but last I heard they're still using a derivative of the system I helped build there. The USB controller died at some point and was replaced with an Arduino, and I think some new summer students have tweaked this and that, but best I can tell the system is still mostly alive (even if it's been through a Ship of Theseus transformation). The results were really cool too! You could watch the insect successfully navigate around obstacles by changing the timing differential between wingbeats! Anyway... I'm happy to chat about this more if anyone's interested, with the caveat that I'm not a biologist and was way more focused on the EE and CS aspects of doing the measurements. I do recall that there was a particular neuron that was being investigated (the Descending Contralateral Motion Detector - DCMD) in our lab, but don't know a whole lot more than that.
foxyv 8y ago
It would be pretty unfortunate if suddenly my car was attracted to bright lights and bug zappers. /joke/

5 points by vezycash 8y ago

That's kind of cool. But I hope we wouldn't be satisfied with a 91% collision avoidance rate.

Ooooh! Story time! In 2005 and 2006, I worked in a biology lab that worked on similar things as Dr. Rind's lab. I don't recall if we collaborated directly, but there was definitely a lot of overlap and I think we shared papers back and forth. When I arrived, the biologists had rigged up a pretty amazing setup. They had a system that was measuring locust wing-muscle activations, and feeding that in through a joystick port to Descent (yes, the game). This was a more closed-loop version of the "watching Star Wars" stuff talked about in the article. The system had a couple flaws, though. The biggest being that it was being projected onto a hemispherical dome that the locust sat in, but the scene wasn't perspective-corrected for the dome. It still worked, sort of though; you could watch the insect successfully navigate around obstacles, but it was pretty crude. Over the two summers (I was a student at the time), I did the majority of development on a more controlled and perspective-corrected system for them. It included a simple scripting language (since I had, you know, just finished my compilers course) for the scientists to describe specific scenes to project for the insect, instead of just using the pre-made Descent levels. It did some really cool projecting mapping, such that, given a known position of the insect's eyes relative to the hemispherical screen, we could project perspective-correct 3D imagery. It had a USB device for more predictable input than the joystick port. And it included time synchronization signals that could be fed back into the muscle-impulse-capture hardware to more easily correlate events in the 3D scene with muscle events. I haven't kept in touch with the lab much, but last I heard they're still using a derivative of the system I helped build there. The USB controller died at some point and was replaced with an Arduino, and I think some new summer students have tweaked this and that, but best I can tell the system is still mostly alive (even if it's been through a Ship of Theseus transformation). The results were really cool too! You could watch the insect successfully navigate around obstacles by changing the timing differential between wingbeats! Anyway... I'm happy to chat about this more if anyone's interested, with the caveat that I'm not a biologist and was way more focused on the EE and CS aspects of doing the measurements. I do recall that there was a particular neuron that was being investigated (the Descending Contralateral Motion Detector - DCMD) in our lab, but don't know a whole lot more than that.

foxyv 8y ago

It would be pretty unfortunate if suddenly my car was attracted to bright lights and bug zappers. /joke/