Thursday, June 20, 2013

Post Lecture Recap: Biorobot (R)evolution

Last night, I traveled down to the beautiful Bell House, in the Gowanus section of Brooklyn, for the latest lecture sponsored by the Secret Science Club. This month's lecture featured Dr John Long of Vassar College. Dr Long is a professor of biology and the head of the Interdisciplinary Robotics Research Laboratory. His work involves biorobotics, efforts to "evolve" robots in order to learn about life.

Robots are useful for performing dull, dirty, and dangerous jobs, the sort of jobs humans should not have to perform. Mercedes and Lexus have created integrated driving systems that employ radar and ultrasound to increase safety by correcting for driver errors. The innovations of the integrated safety systems were developed as part of a project to develop driverless cars. The Google car is a prime example of a prototypical driverless car. Dr Long also mentioned the use of drones as robotic warriors, and touched on a publicity stunt in the UK in which an octocopter delivered pizzas for Dominos.

Dr Long made especial mention of the Roomba, which transitioned to a discussion of the "android fallacy"- most robots do not look like humans. He contrasted a 1960's image of a house cleaning robot with the 21st century reality of a house-cleaning robot.

The next topic of discussion was about what a robot is. A robot exhibits constructive action and "mental" and physical agency that does not result from a biological source. Robot behavior, actions, and movements are sensor guided decisions. A computer produces the communication between the robot's sensors and its mechanisms, and can also communicate with other computers. Roboticist Rodney Brooks, a proponent of the "actionist" school of robotics, famously opined that we should stop making humanoid robots, and should emulate "simpler" organisms, such as insects. The Roomba, mimics an insect-level intelligence and behaves like a bug. One motto used by Lockheed Martin's Skunk Works was "start simple first", an expression akin to the design principle K.I.S.S.

Dr Long's lab is concerned with using robotic models in order to understand vertebrate evolution. He gave a brief overview of the milieu in which the vertebrates first evolved. One of the most famous fossil beds from the time prior to the evolution of the vertebrates is the Burgess Shale, which preserves remains of such famous organisms as Hallucigenia and Opabinia (Dr Long showed a slide featuring a painting of a "school" of Opabinia that he was especially pleased with, but I can't find the image online- suffice it to say, I thought it looked like "Opabinia Jesus" addressing his disciples). One of the earliest vertebrate fossils is the genus Haikouichthys, the remains of which were discovered in China. Haikouichthys was approximately 1 inch (2.5 centimeters) in length. The fossils are especially good because soft tissues were preserved in the fine-grained sediments. Dr Long gave a very terse, hilarious take on the difference between invertebrates and vertebrates: invertebrates are "CRUNCH, SQUISH" organisms while vertebrates are "SQUISH, CRUNCH" organisms. For a better take on vertebrates, vertebrates have an internal support structured called a notochord and paired sense organs (note- with certain exceptions). One sensory organ characteristic of vertebrates is the lateral line, which senses movements and changes in pressure around an organism. The closest analog that humans have to the lateral line is the inner ear, which plays a role in balance.

Dr Long then proceded to discuss his use of robots to model vertebrate evolution. To model something in order to understand it, one has to model simple things first. To study vertebrate evolution, a biorobot, a model of an animal, needed to be built. The model had a series of assumptions built into it, it had to represent a biological target. For his studies, Dr Long decided to model Drepanaspis, a skillet-shaped 400 million year old jawless fish. The target was "simplified", but the functional needs of the model were similar to those of a living organism, and the basic structures were similar. The resultant skillet-shaped swimming robot was dubbed "Tadro, a portmanteau word composed of "tadpole" and "robot". The autonomous "tadro" had light sensors on the front to model feeding behavior (since the base of the marine food chain is phytoplankton, the "feeding" behavior involved heading toward a light source) and "predator" detecting sensors on the side to mimic the lateral line. Swimming was accomplished by a flexible "tail" filled with a hydrogel which Dr Long drolly noted was not edible. The notochord is an ancestral condition, while the vertebrate spine evolved in order to provide increased stiffness. The stiffness of the tail could be adjusted by adding ring-shaped stiffeners, which function as biomimetic "vertebrae".

A simple definition of evolution is a change in the genetics of a population from one generation to the next- certain organisms have more descendants which survive. It is important to note that populations evolve, not individuals.

In order to model evolution, the tadros shared the tank with a "predatory" robot dubbed a "tadiator" (a portmanteau of "tadro" and "gladiator"). Animals can detect predators and override their feeding behavior in order to escape them (Dr Long contrasted this ability to override feeding behavior with the behavior of Augustus Gloop). The tadros, like fish, were able to stop "eating" and start fleeing, the artificial intelligence was able to switch behavior. When fish need to evade a predator, they use a maneuver known as a fast start, in which the spine forms a "C" shape before bursting into a rapid acceleration. In the lab the "tadiator" acted as an agent of "natural" selection for improved feeding and fleeing in order to determine if "evolving" vertebrae would improve these behaviors. In essence, the robots would transition from having notochords to having vertebral columns.

In the lab, six tadros would share the tank with one tadiator. The tadiator, unlike the tadros, would not "evolve", it only serves to hunt the prey robots. Variation among the robots included size of the tail fin, rigidity of the "notochord" (number of "vertebrae") and sensitivity of the "lateral line". Out of the six tadros, the top three (rated by their ability to stay near the light and to avoid tadiator) were allowed to "breed", their variations would be incorporated into the next generation of robots. The best performing tadro would produce six sets of "genes", while the second would produce four, and the third would produce two. The "genes" would be randomly combined to simulate mating, and the next generation of robots would be altered to conform to the "breeding".

Over the course of several generations, the number of vertebrae "evolved" would stabilize after some fluctuation over the course of the first couple of generations.

Here's a good interview with Dr Long, who recently released a book titled Darwin’s Devices: What Evolving Robots Can Teach Us About the History of Life and the Future of Technology (for the record, I did not buy a copy of the book, having spent my money on beer, but I plan on buying a copy when it comes out in paperback). Here is a video of Dr Long in action, which is a good approximation of Tuesday's lecture (be sure to drink a lot of beer while you watch it. Opabina Jesus makes an appearance at 2:56:

In the course of the Q&A, some bastard in the audience asked Dr Long if he had modeled the exoskeletons of such swimming invertebrates as crustaceans, and if similar genetic factors were involved in the evolution of rigid exoskeletons and rigid endoskeletons (hox genes are involved in the development of both). After the lecture, when said bastard went up to Dr Long to thank him for his excellent lecture, he was pleased to be dubbed "crustacean guy" by Dr Long.

In short, Dr John Long of Vassar College is a hell of a good guy, his lecture was excellent, and you should definitely buy his book provided you haven't spent all of your cash on beer. The presentation was yet another excellent effort on the part of the excellent Secret Science Club. As an added bonus, the third deity in the Secret Science Club pantheon, Michael Crewdson, who co-authored the amazing Carnivorous Nights: On the Trail of the Tasmanian Tiger and the indispensable Wild New York with Margaret Mittelbach, returned to his native Brooklyn from his current Australian home. It's been a while since he's been back, but his return was most triumphant. All told, it was an excellent night, even with the "fast track" construction on the "R Train", which resulted in the bastard not getting home until after 3AM (which partially explains why this post is tardy).


Smut Clyde said...

I see no allusion to the sexiness of robots and I call Shenanigans. What have you done with the real BBBB?

paired sense organs (note- with certain exceptions)
Pineal gland, sir.

Any mention of Braitenberg vehicles?

ifthethunderdontgetya™³²®© said...

N and R trains: Never and Rare

Big Bad Bald Bastard said...

I see no allusion to the sexiness of robots and I call Shenanigans. What have you done with the real BBBB?

Damn, I was going to post that video, but this post, late as it was, was a rush job.

N and R trains: Never and Rare

Thunder knows the flavor!