On May 20, 2008, at 12:24 AM, Alexis Lussier Desbiens wrote:

I like the idea of the large robot, although I think it might require a fair amount of work to make it behave realistically. I would rather spend it on the plane itself but let talk tomorrow about that and how it can help us avoiding deadly crashed using the Adept!

To see if the Adept idea is reasonable, I would need to first find out what linear and angular velocities and accelerations the plane experiences in the last stages of flight. If they are within the capabilities of the Adept it should not be hard to set up a safe system with a sort of "mechanical fuse" in which, for example, the wall is very light and easily knocked over in case of any crash. %ENDCOMMENT%

I was thinking about more low end "experiments": using a soft wall first to absorb some of the impact if something goes wrong, a set of small lines to prevent the plane from crashing on the ground, throwing the plane by hand to test the spines, building a low speed-ultralight plane (reduce the momentum)! And as a first demonstration we can do the hover with a slight forward velocity to land on the wall... that will work!

These are good ideas too. Alexis -- is it worth our getting our high speed (500 fps) video going to have a closer look at what the plane is exactly doing?

One thing about the Adept is that if it can achieve a particular trajectory, it will be highly repeatable and the trajectory will be exactly known. So it could be possible to run many tests in a short time with a high peecentage of useful data.

Mark %ENDCOMMENT%

---

From

alexis.lussier.desbiens@gmail.com

Subject

Re: Notes on your replies.

Date

May 20, 2008 12:24:54 AM PDT

To

cutkosky@stanford.edu

Cc

paul@coe.drexel.edu

I like the idea of the large robot, although I think it might require a fair amount of work to make it behave realistically. I would rather spend it on the plane itself but let talk tomorrow about that and how it can help us avoiding deadly crashed using the Adept!

I was thinking about more low end "experiments": using a soft wall first to absorb some of the impact if something goes wrong, a set of small lines to prevent the plane from crashing on the ground, throwing the plane by hand to test the spines, building a low speed-ultralight plane (reduce the momentum)! And as a first demonstration we can do the hover with a slight forward velocity to land on the wall... that will work!

Alexis

---

Begin forwarded message:

From: mark cutkosky <cutkosky@stanford.edu> Date: May 19, 2008 11:11:42 PM PDT To: "Paul Oh" <paul@coe.drexel.edu> Cc: "Alexis Lussier Desbiens" <alexis.lussier.desbiens@gmail.com> Subject: Notes on your replies.

Paul - thanks for your notes. I had not thought about the blimp idea. Very interesting. As I think about it more, I wonder about a couple of issues:

1. A blimp doesn't really need to perch -- I realize this is not the point, but it may make for a more confusing story...

2. Blimp dynamics, although well understood, are perhaps very different from those of a fixed-wing aircraft. How transferable will results be? There is also the practical issue of getting the bottom of the blimp close enough to a wall...

Nonetheless, your point about having a good near-term solution -- so that work can proceed in parallel on plane control and spine deployment/engagement -- is well taken.

I was wondering if we could use the large robot in my lab to execute series of trajectories. (The robot is a high-speed, high acceleration version of the Adept One, with 5 axes.) It could probably recreate most airplane velocity trajectories with reasonable accuracy. We could vary the trajectory by various amounts to ascertain sensitivity and get an idea of the "pre-image" size. The one thing that would be very different is the reflected inertia of the Adept end effector versus a plane. It would likely be necessary to make a very light suspension to isolate the inertia of the Adept from the spines and landing mechanism. One could even attach a plane to the Adept end-effector, via a spring-loaded gimbal or ball joint so that the various sensors on the plane were in use. Thus, the Adept would become a bit like a kid playing with a toy airplane -- except that it would be capable of many repetitions at high speed and high accuracy.

I think, although I'm not entirely sure, that this setup could provide useful data. We did something similar in the 1st year of RiSE? . Boston Dynamics had not yet built the RiSE? robot so we made a test track with a single leg on rails for testing spines, claws, etc. The dynamics were necessarily somewhat different from those of the untethered robot but we learned useful things about approach and departure angles, peak forces, etc. for the claws and spines.

Alexis - any thoughts? You've been very quiet...

Mark

---

From

paul@coe.drexel.edu

Subject

RE: Some initial proposal thoughts (longish message)

Date

May 19, 2008 6:34:41 PM PDT

To

cutkosky@stanford.edu, alexis.lussier.desbiens@gmail.com

Mark and Alexis,

Reading your thoughts, I kept thinking of the testing and evaluation (T&E), the verification and validation (V&V) steps. It may make sense to employ something like a blimp for early testing. Blimp dynamics are quite well-understood. Also, T&E using blimps will not result in disastrous crashes or risk damaging test areas. Just a thought...

-paul

---

Original Message----- From: mark cutkosky [mailto:cutkosky@stanford.edu] Sent: Sunday, May 18, 2008 9:46 PM To: Alexis Lussier Desbiens Cc: mark cutkosky; Paul Oh Subject: Some initial proposal thoughts (longish message)

Alexis -- I would like to start drafting a proposal on the vertical surface perching. I will set up a folder on BDML where we can collect things. The main sections need to include:

Motivation -- includes statement of broader impact * related work

Technical approach * Research methods * Plan of experiments

(If the proposal goes to NSF or DoD? there will be other more specific requirements... )

Paul -- I would like to keep the details somewhat confidential for now -- at least until we're at the point that we've either written a proposal or been able to write a results paper that describes a first success.

= I have a general idea of what we want to do. What we need to work on is formalizing the process so that it is easily understood by somebody who doesn't already have a bunch of experience in microspine design and deployment.

Some areas that need work (Alexis: please reply to this email with any initial thoughts you have):

1. Landing on a vertical wall: the final 20 centimeters of flight. How to frame the sensing, control research issues -- what experiments, what hardware, what methods and what tests. What metrics of success to guide the work. Discuss absolute position and/or velocity sensing (e.g. ultrasonic, laser range, optical flow) in addition to inertial sensing. At short distances it should be possible to get excellent orientation and velocity information (Danko et al. 2005; Green et al. 2006).

2. How to frame the problem of designing and controlling the mechanisms for deploying, attaching and loading spines.

The Spinybot IJRR paper is good for describing the relationships among spine tip radius, strength, asperity size and the probability of engaging asperities per unit area of wall. However, it doesn't say much about mechanism design. We need something that sounds better than "enlightened trial and error" (which to be honest, is more or less what we used on Spinybot). Compliant underactuated mechanism modeling and analysis is easy; synthesis is hard.

One way to frame the problem is to decompose it into (i) determining a desired force/motion trajectory for the spines using an assumed, perhaps probabilistic, trajectory of the vehicle with respect to the wall. (I assume the plane will have significant variations with respect to any nominal trajectory.) Then, (ii) synthesize very lightweight, compliant mechanisms that will deploy and engage the spines given the plane trajectory. Today's mechanism synthesis programs (e.g. Khota, Lu) are not quite equal to this task. Instead, a variety of configurations will be examined using dynamic numerical models and likely candidates will be tested and refined. High speed video and force measurements (multi-axis force plate on the wall) will provide information for tests and refinement.

Conversely, maybe the problem should be inverted -- given constraints on feasible spine engagement trajectories, first define the compliance matrices associated with the spines and establish a range of possible spine approach and attachment trajectories that will work reliably for a given set of spline compliances and spine geometries. Use this space of trajectories as a "pre-image", in the sense of robot motion planning (e.g., Latombe). Then the aircraft control problem is to get the plane within the pre-image for which there is a high probability of spine attachment.

The former approach makes more sense if the space of possible airplane trajectories provides the more severe constraint; the latter approach makes more sense if the spine engagement mechanism and its trajectory presents the more severe constraint. At present it is hard to answer this question. In the case of a climbing robot like Spinybot, the robot had only a very simple (essentially one dimensional) trajectory parallel to the wall so the problem of designing the mechanisms for engaging the spines was much less open ended.

3. Once the first set of spines has engaged, a second, opposing set of spines is engaged and an internal force is generated between them. This second engagement problem should be easier because there are now fewer degrees of freedom. It is possible that the initial spines will loose their grip during this process but (i) this can be minimized by maintaining a small normal force into the wall and (ii) if the mechanism is designed well, they will reengage quickly. Once opposed spines are engaged, applying a significant internal force in the plane of the wall will ensure that the vehicle clings tenaciously. With sufficient force, it could even ride out a storm.

4. Take-off is much easier. It suffices to reverse the internal force to achieve disengagement. If there is clearance, the propellors can be brought up to speed prior to release. Also, there is the opportunity for jump-assisted take off, using potential energy stored in the legs.

---