Hacker News Reader

Comment by hawkharris | Hacker News Reader

A few years ago I enrolled in a four-month investigative journalism program. My challenge was to find an interesting character and embed myself with the person, writing about his or her life and career.

I ended up running into an engineer from Boston Dynamics and spent four months writing about his unique approach to solving problems.

The engineer, Jon, who worked on Big Dog as well as Rise, was one of the coolest and most inspirational people I've met. He explained how he struggled with dyslexia as a kid, and engineering, being more right brain-oriented, was a vehicle that allowed him to succeed despite his disability.

In fact, he later took a huge salary cut so he could become a high school teacher and teach engineering to kids, especially those with dyslexia.

But getting back to the issue at hand, I really enjoyed learning about Boston Dynamics during the four months I spent working on the story, and I think Google made a smart decision.

View on HN

do you happen to mean jon clark? i helped work with him on RiSE :) can't believe it's been almost 8 years now. i wasn't working with the robotics directly, though, but studying animal locomotion (measuring foot forces, CM, roll/pitch/yaw ...). good times

Do you happen to have any reports or theses about the animal locomotion studies? I'm currently working on a similar topic (https://github.com/ivoflipse/Pawlabeling/) and its always good to learn from past experiences.

sure! most of my publications were centered around how robots should climb vertically, but i am also familiar with horizontal and also on various substrates (sand, mesh, etc). we studied the animals "steady state" locomotion (such sprinting as fast as you can without thinking about your steps). this paper is the culmination of many hours spent with cockroaches, geckos, home-made 3d force sensors, high speed cameras, matlab, and c++ simulations :)

http://www.roboticsproceedings.org/rss03/p02.html

let me know if you have any questions or would like to talk about other animals, substrates, and/or differences between horizontal and vertical locomotion.

That looks really interesting! I'm currently trying to cook up a reliable way to label my paw prints (like here: http://stackoverflow.com/questions/4502656/how-to-sort-my-pa...).

I'm wondering if modeling/simulating quadruped locomotion would help make this task a lot easier. Instead of trying some machine learning/computer vision on the pressure data, reason about what kind of walking pattern brought forth the results. Like a paw can't be in two places at the same time and other heuristics like that.

Any suggestions for literature I should read up on for horizontal locomotion? :-)

cool. what specifically are you trying to study?

there's no substitute for the actual foot force data. i would collect this directly, in 3d (not just z-direction -- which it seems like you are currently doing), to understand the basics and then later involve simulations to iterate different designs that produce the desired motion and forces. i built a force plate that measures forces in 3d with piezoelectric sensors. it shouldn't be as hard with a feline because their feet have a reasonable distance between them (try getting only one of a cockroach's feet to be touching the plate at a given time to measure what forces it produces!). the reason why i have this opinion is that the most interesting part of comparing how animals run versus climb are the lateral forces. another interesting thing is that the number of legs is irrelevant (well except maybe monopod) because when studying two, four, six or 44 (centipede -- man those things are scary to handle!) legs, the forces reduce to a basic pattern we all share in common. when a centipede is sprinting, only three of its legs are on the ground at any time, so it's forces look identical to a cockroach or any other hexapod. when you have less than 6 legs, then you can observe the effective 6 legs by analyzing various portions of their phase. for example, when you walk (bipedal), when your foot is out in front of your center of mass, the forces generated are like the front feet of a hexapod (deceleration only). when you foot is near your CM, it generates forces like the middle legs of a hexapod (deceleration followed by acceleration). and when your foot is behind your CM, it generates forces like the rear feet of a hexapod (acceleration only).

there's more i want to say, but i'll save it for a subsequent comment :)

as for additional reading for you, i would take a look at Robert J. Full's research. there are a couple horizontal locomotion papers cited in the my paper linked in the prior comment that have phrases like "on the horizontal plane" or "on land" in the titles.

also, here's all the papers from his lab: http://www.polypedal.com/?page_id=163

and here's a link to one to get you started: http://www.polypedal.com/Images/PolypedalPublications/62_Dyn...

I'm helping veterinary scientists with their gait analysis, basically by porting the calculations used for pressure measurements (so no 3D forces) on humans and solving problems like detecting the paws along the way.

An example of what this gets used for are things like lameness detection or diagnosing disorders. Or it could be used to evaluate the effectiveness of the treatment.

While the pressure plate doesn't measure 3D forces, it does have the advantage of giving you information about the distribution, which I expect at least has strong correlations with 3D forces. Obviously you need animals of a certain size to get good data (or systems with very small sensors), so you wouldn't use it for gecko's or even really small dogs.

Perhaps I should try calculating the distance to the body's center of mass, to get an idea how well this distinguishes the different paws. The problem I'm having so far is that the quality of the measurements isn't always optimal and inferring which paws are making contact is difficult if you don't see paws from both sides (for quadrupeds at least).

I'll be sure to check out the research you pointed to!

Jon Clark is a professor at FSU now, and he's still collaborating with some ex-RiSE folks like Mark Cutkosky (my PhD advisor) and Dan Koditschek.

The robotics expertise that Google is assembling is quite varied, which is really interesting. They already have stellar computer vision and machine learning people on the self-driving car team, and their earlier acquisitions cover other software disciplines like whole-arm kinematics (Bot & Dolly) and vision (Industrial Perception) as well as hardware like manipulation/grippers (Meka and Redwood) and actuator design (Schaft).

The Boston Dynamics team is probably the best concentrations of specialists in dynamic legged locomotion, especially by studying animals for inspiration. Marc Raibert, who started BDI, did stuff in the 80s that still seems incredible today.

http://www.ai.mit.edu/projects/leglab/old-leglab/robots/robo...

I'm excited to see what they can do with the flexibility and funding they'll get at Google. I believe their previous funding was primarily grants from DARPA, which means lots of grant proposal writing, narrowly focused research goals, quarterly progress reports, etc.

I did my undergrad with Jon Clark! Great guy. It's cool to see a few bio-robotics people on HN

The guy I wrote about was Jon Amory, but Jon Clark sounds like a cool character as well. Here's the story, which includes some photos and videos. You might recognize the other Jon because I think he also spent a lot of time working on RISE. :)

http://codynedromano.wordpress.com/2011/04/08/vertices/

I just wanted to say you're a really talented writer! :)

Thanks! That's very nice of you to say. :) It helps a lot when you stumble upon an interesting subject and character.

Interesting. Do you have a link to the story?

Hey, thanks for your interest. :) Here's a link: http://codynedromano.wordpress.com/2011/04/08/vertices/

Reply on news.ycombinator.com