[Katie Byl]
  Katie Byl    [KAY-tee BILL]    katiebyl@ece.ucsb.edu    ECE594d
My UCSB homepage (with link to online calendar) is here: http://www.ece.ucsb.edu/~katiebyl

As of January, 2010, I am at the

...as a member of the faculty in the ECE Department. My research is in Controls: specifically, toward improved robot dynamics for reliable and effective locomotion and manipulation, and more generally in developing methodologies (e.g., learning algorithms) to create agile autonomous systems.

In 2009, I was a postdoc, working on control of tiny, dynamic, flying robots at Harvard; one application for such artificial insects is to create RoboBees as a hedge against hive collapse. As a PhD student, I studied legged robots at MIT; some video from work with LittleDog appears later on this page.

The Byl family have recently moved from the "fair city of Cambridge, MA" to fair-weathered Santa Barbara, CA. Our other big news this past year is the birth of our son, Pieter Byl.

You can see me (and labmates) doing some serious trash talking on robotics in a fun mockumentary on the Goblin Man (go to Part 3: exhibition). Note, LittleDog gets a starring role, too.

The funnest part of my PhD research was developing these dynamic motions for LittleDog. And yes, LittleDog is in fact the smaller sibling of the BigDog robot. Both are products of Boston Dynamics.
[Katie Byl]
   [cool video at left...]

I have a notorious past. I was a professional gambler for several years on the infamous MIT Blackjack Team. My name was Katie Lilienkamp back then, and the "team" was actually a series of variously-named LLCs [warning: That last, wikipedia-page link is pretty unbalanced. Maybe some day the many-score various past-teammates and I will attain activation energy to retify that... but I would not hold my breath for that eventuality.]

Side note: Successful gamblers and roboticists have two things in common: getting paid to play, and a proclivity to work late-night hours in windowless spaces. If you are weighing these two career options personally, however, do not belittle the fact that your average robot lab affords a much more smoke-free daily working environment.

I've TA'd an epic number of times at MIT (see my CV!) . If you had me as a TA, please do drop me a line! I am curious what adventures and turns your life has taken (even if you didn't end up studying controls).
    [Katie and Pieter Byl]

Selected Publications and Presentations

List to be updated...
Metastable Walking on Stochastically Rough Terrain, RSS 2008 (Katie Byl and Russ Tedrake) [movie]
Approximate Optimal Control of the Compass Gait on Rough Terrain, ICRA 2008 (Katie Byl and Russ Tedrake) [movie]
Control of the Compass Gait on Rough Terrain , Dynamic Walking (DW) 2008 (Katie Byl and Russ Tedrake) [pdf of ppt presentation]

Lab Experiences for Teaching Undergraduate Dynamics - M.S. thesis, Katie Byl (nee Lilienkamp)
A Simulink-Driven Dynamic Signal Analyzer - B.S. thesis, Katie Byl (nee Lilienkamp)
Dynamic Signal Analyzer for dSPACE - Quick documentation of the DSA developed for my Bachelor's thesis

More Cool Robot Videos

i.e., the good stuff is just ahead... Happy hunting!

LittleDog Quadruped Robot

- LittleDog Collage   76MB
- DARPA Phase 2 Highlights   76MB
- Walking on pegs. (aka Karate Kid) [this video above is sped up 3x!]   4.8MB
- Reliable double-support maneuvers for a high-impedance quadruped   5MB
- Gymnastic vault-climb of gap and tall
box obstacle, with pacing at end
- One-legged flying push-off during gap crossing   12MB
- Dynamically climbing up onto terrain G. See also multiple consecutive trials: 1y[10MB] 2n[2.9MB] 3y[9.8MB] 4y[9.3MB] 5y[7.8MB] 6y[9.2MB] 7n[2.6MB]   16MB
- Valiant shove at end during gap crossing   12MB
- Jersey barrier with dramatic death-dive to reach goal ASAP   4.0MB
- Recovery push-up for use if we do not reach the "goal" after a dramatic death-dive   3.7MB
- Headstand?! We found the dog could execute this useful - but quite odd! - strategy to climb a step. The "hop" avoids kinematic collisions to untangle the front legs from the back. Note cool recovery push-up after failed deathdive-to-goal, too.   4.8MB
- Totally dynamic stair climbing (Run 3). This was so fun to code up! It ends with the headstand sequence from above. Looks like thrashing, but check out the repeatability by looking at two other runs of this code: Run 1  Run 2   4.8MB
- Gray modular terrain at nearly 3 times DARPA Phase 2 metric speed.   5.1MB
- Steps, with jumping ...quite graceful.   10MB
- Steps, with sliding, is actually just as fast.   11MB
- Fast walking with a double-support phase   3.7MB
- Molded rock terrain. These are manufactured rocks - intended to hold your spare house key!   15MB
- The Uncanny Valley for DOGS?. The real dog watching here is Murphy. Here's a wikipedia link to find out what the uncanny valley is in robotics.   12MB
- Bounding gracefully is tough to do with such a high-impedance robot. This is definitely where a little more (SLIP-esque) compliance would go a long way...   16MB
- Bounding across a barrier gracefully is even more difficult to do! Compare various attempts to get over the jersey barrier in a single, smooth motion. Take 2 Take 3 Take 4 Take 5 Take 6 Take 7 Take 8 It never works, but it also fails with an impressive LACK of repeatability! (ugh.)   1.4MB
- Phase 1 results were not quite so dynamic, but a lot of effort (labwide) went into getting these early results. Another example of Phase 1 terrain.   16MB
- Unintended front flip. Outtakes are fun. Here is video from early debugging of the gaps terrain.   2.8MB
- Unintended BACK flip. Outtakes are fun (did I mention??). Here is video from early debugging on the jersey barrier.   7.0MB


Compass Gait Biped Model

- One-step policy, Rough Terrain   6.1MB
- One-step policy, Rough Terrain with Gaps   8.0MB
- Sensitivity to particular one-step policy,
Rough Terrain with Gaps, but falls!
- Chaotic walking on flat terrain when executing
optimal policy for rough, blind terrain ahead!

More video coming soon...