Legged Robots

I worked on legged robots in college. The CAD models are all viewable in the browser via "Onshape document" links.

AMBER 1

AMBER 1 was the first thing I worked on in college.  It was a lot of fun trying to figure out ways to get tiny (4mm) motor shafts to transmit constantly reversing, high torque without shredding whatever machinable material was available. We eventually settled on chains and sprockets, the set screws of which would constantly back out regardless of whatever thread locker we could find.  Ultimately, just a small tack with a TIG torch settled it once and for all. It wasn't pretty, but it was enough to verify the reduced-parameter optimization-based gait generation methods that the lab was working on at the time. They would expand out into wider scopes in short order.

 AMBER 2

Making AMBER 2 is probably the most fun I had during a summer in college.  I had worked initially with Jordan Lack to make AMBER 1, but he was in his master's program and therefore was eligible to go to work on the DARPA robotics challenge at Johnson Space Center, and I had to make this work from the design standpoint mostly on my own. It was a rough lesson in the realities of research grants. My advisor, Aaron Ames, was adamant that it be close to human form so that he could get funding. "Make it cute. I have to sell this." And he was absolutely right.

AMBER 3

AMBER 2.5-3 (I'm not sure when the name was settled) represented the opportunity to make something of similar kinematic complexity to AMBER 2 at realistic humanoid scale. I worked with Eric Ambrose on this during the summer of 2014, remotely, with a baby in a carrier on my chest half the time. We used the typical "Dropbox as version control" method. There were some changes after the lab's relocation, so this isn't perfectly representative of where I left off for my next robot, however, it is very close.

Series Elastic Actuator Spring

One aspect of the full-size Amber robot that was planned but deemed unnecessary was series elasticity. The concept borrowed heavily from the series elastic actuators in NASA's Valkyrie project. I'd eventually get to know several of the members of that team at Novium Designs, which was a bit of a fanboy moment. Their work was fantastic. Having played with the spring design a bit, I eventually worked with Chris McQuinn who had designed that very part. 

 Unnamed Raibert-Style Hopper

I made this hopper for my master's thesis, but due to Aaron Ames' getting recruited twice in two years (once to Georgia Tech, then to Caltech), none of the timing worked out to properly fund it and transfer all the motors and components between the different universities.

I took a page from Dr. Jonathan Hurst (now of Agility Robotics) and used a cable and pulley system, as in MABEL. MABEL's performance is impressive even today, with Boston Dynamics making Superbowl-upstaging commercials. That robot had a brutal period of system identification, which was partly due to the cables and pulleys, but I thought that could be minimized by increasing the pulley diameters and using smaller cables under lower bending stresses. Using motors with large air gap diameters and a low speed reduction would have minimized impedance and impact stresses in the fashion of Dr. Sangbae Kim's Cheetah, eliminating the need for compliant springy elements to keep from shaking itself to pieces. On paper, it was a beast.

I had several feet designed for different scenarios but the CAD files seem to be lost to the woes of Dropbox version control.

Analyses

 It was outrageously capable, and would have been able to jump considerably higher than it stood from a theoretical "flat footed" stance.  I verified the design in simulation by using Lagrange multipliers to pull out the forces at the joints, and then using those as the basis for FEA on the individual parts. The red dots here represent forces pulled out of the reset map's perfectly plastic impact assumptions.

Plotting the torque/speed trajectories against the motor's capabilities shows that they're capable of higher instantaneous torques than the system requires, although the RMS values (the yellow star) were getting close to the manufacturer published specs.

AMBER Lab Test Fixtures

Boom

I designed AMBER Lab's boom setup  to support a wide variety of robots with no adjustment or fuss by using a four bar linkage. In theory, that would have let the lab test on a wider range of stairs and other obstacles without edging into gravity being a nonlinear factor, but those gait tests never materialized. In order to avoid bogging down smaller robots like AMBER 2 it needed a counterweight, the rotational inertia of which I minimized by keeping it as close to the axis of rotation as possible.

Overhead Track

This linear test track was a copy of something Stanford Research Institute put together at their facility for the DURUS/PROXI project. As the lab was getting that robot to run experiments, we ended up getting money to put in a test track. A&M put some channel strut in the ceiling, but there wasn't a right angle or straight line to be found. I managed to shim and hammer the whole thing into useful, straight lines over the course of a week or so.