Dan Aukes' box of goodies

For discussion of hand designs.
Posts: 4
Joined: Mon May 04, 2015 5:12 pm

Dan Aukes' box of goodies

Postby Nick » Sun May 10, 2015 5:04 am

Something like a month ago, this PhD student Dan Aukes from Stanford came to Cornell and gave a really interesting talk about some research of his relating to robotic hands and building complex layered materials on the cheap. I meant to get him on an R&D call sometime, but that fell through the cracks so now I'm posting his most relevant research here, with summaries. (pdfs generously provided by Cornell's IEEE subscription :P)

1. Selectively Compliant Underactuated Hand for Mobile Manipulation: http://www.mediafire.com/view/cznt9gpu5 ... lation.pdf

2. Varying spring preloads to select grasp strategies in an adaptive hand: http://www.mediafire.com/view/vnol2eu46 ... eloads.pdf

3. A Robust, Low-Cost and Low-Noise Artificial Skin for Human-Friendly Robots: http://www.mediafire.com/view/sys88c214 ... iendly.pdf

1. This is what really caught my attention in his talk at Cornell. The paper is about a gripper which has the ability to lock its joints selectively, and how locking/unlocking joints at different times changes the way it grips different objects. It's awesome, but locking joints like that is expensive and not really in e-NABLE's budget.

The bit I'd like us to pluck from this paper is the simulations he ran. Essentially, he used a rigid body simulation of his hand (Blender, 3ds max, etc are all capable of this. Plus blender has python scripting) to see where a given object would go if it started in a certain place, and the hand closed. Then he did this with tons of starting positions and traced all their paths to create a kind of vector field for "where does this hand send this object?" The resulting field has poles where an object is stably gripped. Then you just trace the region of the field which eventually flows to those poles, and you can say that if the object starts anywhere inside that region, the hand will grab it successfully.

Essentially, it's a quantitative way to validate our hand designs much more thoroughly than we could do with a physical set of test objects.

2. This one's about fingers like ours which have only one actuated degree of freedom, but applies some of the ideas of number 1. The first paper used locked joints to maximize the size of the "successful grip" region, while this one uses different spring tensions on the joints to control the order they close in. I haven't read it in much depth, but it sounds readily applicable to our hands, particularly the new variety with rubber bands instead of elastic cords.

3. This one's on a different topic than the other two, but also awesome and applicable to e-NABLE reasonably quickly. The team created an array (4x4) of touch sensors in a flexible skin for use on the surface of a robot. Multiplexed arrays like theirs are an awesome idea for the far future, but of more immediate interest is how they built the individual cells, and why. The shielding layer is moderately interesting, but the coolest bit is the idea of a multivibrator based capacitive touch sensor. It's a capacitive pressure sensor that outputs a digital signal (the frequency is related to the pressure), which makes it fantastic at handling noise and working in various environments. It's also dirt cheap, so I think we should really consider implementing these in our designs.

Anyway, that's the gist of Dan Aukes' relevant research. I'd love to see some discussion on here!

Jason M Bryant
Posts: 198
Joined: Mon Feb 16, 2015 1:59 am

Re: Dan Aukes' box of goodies

Postby Jason M Bryant » Sun May 10, 2015 6:29 am

Cool stuff. I'll definitely dig into it later.

I think someone posted something related to number 1 a few months ago. It wasn't this in depth, but I member a summary of results that talked about bending the fingers until the joints met resistance, then bending the next set of fingers. It's why I've been working on finger tensions that mimics that lately. Selectively locking joints may be out of our range, but figuring out the most useful joints to bend first is quite doable.

Return to “Hands”

Who is online

Users browsing this forum: No registered users and 1 guest