Custom mesh jittering in Mujoco environment in OpenAI gym - reinforcement-learning

I've tried modifying the FetchPickAndPlace-v1 OpenAI environment to replace the cube with a pair of scissors. Everything works perfectly except for the fact that my custom mesh seems to jitter a few millimeters in and out of the table every few time steps. I've included a picture mid-jitter below:
As you can see, the scissors are caught mid-way through the surface of the table. How can I prevent this? All I've done is switch out the code for the cube in pick_and_place.xml with the asset related to the scissor mesh. Here's the code of interest:
<body name="object0" pos="0.0 0.0 0.0">
<joint name="object0:joint" type="free" damping="0.01"></joint>
<geom size="0.025 0.025 0.025" mesh="tool0:scissors" condim="3" name="object0" material="tool_mat" class="tool0:matte" mass="2"></geom>
<site name="object0" pos="0 0 0" size="0.02 0.02 0.02" rgba="1 0 0 1" type="sphere"></site>
</body>
I've tried playing around with the coordinates of the position and geometry but to no avail. Any tips? Replacing mesh="tool0:scissors" with type="box" gets rid of the problem entirely but I'm back to square one.

As suggested by Emo Todorov in the MuJoCo forums:
Replace the ground box with a plane and
use MuJoCo 2.0. The latest version of the collision detector
generates multiple contacts between a mesh and a plane, which
results in more stable simulation. But this only works for
plane-mesh, not for box-mesh.
The better solution is to break the mesh into several meshes, and include them as multiple geoms in the same body. Then MuJoCo will construct the convex hull of each sub-mesh, resulting in multiple contact points (even without the special plane mechanism mentioned above) and furthermore it will be a better approximation to the actual object geometry.

Related

How can I use "Interpolated Absolute Discounting" for a bigram model in language modeling?

I want to compare two smoothing methods for a bigram model:
Add-one smoothing
Interpolated Absolute Discounting
For the first method, I found some codes.
def calculate_bigram_probabilty(self, previous_word, word):
bigram_word_probability_numerator = self.bigram_frequencies.get((previous_word, word), 0)
bigram_word_probability_denominator = self.unigram_frequencies.get(previous_word, 0)
if self.smoothing:
bigram_word_probability_numerator += 1
bigram_word_probability_denominator += self.unique__bigram_words
return 0.0 if bigram_word_probability_numerator == 0 or bigram_word_probability_denominator == 0 else float(
bigram_word_probability_numerator) / float(bigram_word_probability_denominator)
However, I found nothing for the second method except for some references for 'KneserNeyProbDist'. However, this is for trigrams!
How can I change my code above to calculate it? The parameters of this method must be estimated from a development-set.
In this answer I just clear up a few things that I just found about your problem, but I can't provide a coded solution.
with KneserNeyProbDist you seem to refer to a python implementation of that problem: https://kite.com/python/docs/nltk.probability.KneserNeyProbDist
There exists an article about Kneser–Ney smoothing on wikipedia: https://en.wikipedia.org/wiki/Kneser%E2%80%93Ney_smoothing
The article above links this tutorial: https://nlp.stanford.edu/~wcmac/papers/20050421-smoothing-tutorial.pdf but this has a small fault on the most important page 29, the clear text is this:
Modified Kneser-Ney
Chen and Goodman introduced modified Kneser-Ney:
Interpolation is used instead of backoff. Uses a separate discount for one- and two-counts instead of a single discount for all counts. Estimates discounts on held-out data instead of using a formula
based on training counts.
Experiments show all three modifications improve performance.
Modified Kneser-Ney consistently had best performance.
Regrettable the modified Version is not explained in that document.
The original documentation by Chen & Goodman luckily is available, the Modified Kneser–Ney smoothing is explained on page 370 of this document: http://u.cs.biu.ac.il/~yogo/courses/mt2013/papers/chen-goodman-99.pdf.
I copy the most important text and formula here as screenshot:
So the Modified Kneser–Ney smoothing now is known and seems being the best solution, just translating the description beside formula in running code is still one step to do.
It might be helpful that below the shown text (above in screenshot) in the original linked document is still some explanation that might help to understand the raw description.

Proj4JS projection definition for rotated pole

I am trying to define a rotated pole projection in Proj4JS where the north pole is now is 48N and 176E. I haven't been able to find any other example of rotated-poles in Proj4JS so I have tried to convert one I found for PROJ.4.
proj4.defs('myProjection', '+proj=ob_tran +o_proj=latlon +o_lon_p=-176 +o_lat_p=48 +lon_0=0 +a=1 +to_meter=0.0174532925199');
That line of JS is run without problem, but when I try to use that projection
proj4('EPSG:4326', 'myProjection', [175, -41]);
I get this error
uncaught exception: myProjection
I've tried replacing the projection definition the one for WGS84 and it works fine, so I believe my use of the function is correct, it's the parameters in that string that I am unsure of.
I think what you want is the so-called Azimuthal Equidistant projection. It's the best choice for measuring true distances radiating away from a center point.
If this is what you're looking for, I asked a similar question a while back over on GIS.SE, and for the coordinate you provided (48N, 176E), you could declare the Proj4js projection definition as so..
Proj4js.defs["CUSTOM:10001"] = "+proj=aeqd +lat_0=48.0 +lon_0=176.0 +x_0=0 +y_0=0 +ellps=WGS84 +datum=WGS84 +units=m +no_defs";
I hope it helps.

game - how can I drag objects (cars with numbers) into targets (start line) AS3.0.?

I am having this problem, where I have several cars, numbers and letters, and need to put 5 cars in the starting places. -random order is ok.
I' having trouble finding in AS3 a way so that the EndX and EndY of each object can be in the starting lines and be considered right no matter the order!
I'm having trouble putting the code here so, heres a titanpad with the code:
this is the code:
being (um, dois, tres, quatro) the movieclip instance name for each numbered car.
https://titanpad.com/42vtnCbvLu
First of all, you could probably benefit by using the distance between two points formula and seeing if the distance is less than a certain value rather than checking in all 4 directions manually:
Math.abs(Math.sqrt((x2-x1)^2 + (y2-y1)^2))
Let the position of the car be (x1,y1) and the start position (x2,y2).
This formula will give you the distance between the two points in any direction, and you could test maybe whether this value is less than your offset.
As for the cars in any order part, I'm interpreting that you have your cars and you want the user to drag them to one of 5 spots, a bit like this:
spot1
spot2
spot3
spot4
spot5
All with respective coordinates. My suggestion would be to have a boolean flag for whether each spot is occupied that stops the program checking whether a car is put there after it has been taken once.
Once all these flags are true, then you can proceed.
Hope this helps.

more minimaler cubism.js horizon chart from json example

Following up on a previous question... I've got my minimal horizon chart example much more minimaler than before ( minimal cubism.js horizon chart example (TypeError: callback is not a function) )
<body>
<div class="mag"></div>
<script type="text/javascript">
var myContext = cubism.context();
var myMetr = myContext.metric(function(start, stop, step, callback) {
d3.json("../json/600s.json.php?t0=" + start/1000 + "&t1=" + stop/1000 + "&ss=" + step/1000, function(er, dt) {
if (!dt) return callback(new Error("unable to load data, or has NaNs"));
callback(null, dt.val);
});
});
var myHoriz = myContext.horizon()
.metric(myMetr);
d3.select(".mag")
.call(myHoriz);
</script>
</body>
The d3.json() bit calls a server side .php that I've written that returns a .json version of my measurements. The .php takes the start, stop, step (which cubism's context.metric() uses) as the t0, t1, and ss items in its http query string and sends back a .json file. The divides by 1000 are because I made my .php expect parameters in s, not ms. And the dt.val is because the actual array of my measurements is in the "val" member of the json output, e.g.
{
"other":"unused members...",
"n":5,
"val":[
22292.078125,
22292.03515625,
22292.005859375,
22292.02734375,
22292.021484375
]
}
The problem is, now that I've got it pared down to (I think) the bare minimum, AND I actually understand all of it instead of just pasting from other examples and hoping for the best (in which scenario, most things I try to change just break things instead of improving them), I need to start adding parameters and functions back to make it visually more useful.
Two problems first of all are, this measurement hovers all day around 22,300, and only varies +/- 10 maybe all day, so the graph is just a solid green rectangle, AND the label just says constantly "22k".
I've fixed the label with .format(d3.format(".3f")) (versus the default .2s which uses SI metric prefixes, thus the "22k" above).
What I can't figure out is how to use either axis, scale, extent, or what, so that this only shows a range of numbers that are relevant to the viewer. I don't actually care about the positive-green and negative-blue and darkening colours aspects of the horizon chart. I just used it as proof-of-concept to get the constantly-shifting window of measurements from my .json data source, but the part I really need to keep is the serverDelay, step, size, and such features of cubism.js that intelligently grab the initial window of data, and incrementally grab more via the .json requests.
So how do I keep the cubism bits I need, but usefully change my all-22300s graph to show the important +/- 10 units?
update re Scott Cameron's suggestion of horizon.extent([22315, 22320])... yes I had tried that and it had zero effect. Other things I've changed so far from "minimal" above...
var myHoriz = myContext.horizon()
.metric(myMetr)
.format(d3.format(".2f"))
.height(100)
.title("base1 (m): ")
.colors(["#08519c", "#006d2c"])
// .extent([22315, 22320]) // no effect with or without this line
;
I was able to improve the graph by using metric.subtract inserting it above the myHoriz line like so: (but it made the numerical label useless now):
var myMetr2 = myMetr.subtract(22315);
var myHoriz = myContext.horizon()
.metric(myMetr2)
.format...(continue as above)
All the examples seem so concise and expressive and work fine verbatim but so many of the tweaks I try to make to them seem to backfire, I'm not sure why that is. And similarly when I refer to the API wiki... maybe 4 out of 5 things I use from the API work immediately, but then I always seem to hit one that seems to have no effect, or breaks the chart completely. I'm not sure I've wrapped my head around how so many of the parameters being passed around are actually functions, for one thing.
Next hurdles after this scale/extent question, will be getting the horizontal time axis back (after having chopped it out to make things more minimal and easier to understand), and switching this from an area-looking graph to more of a line graph.
Anyway, all direction and suggestion appreciated.
Here's the one with the better vertical scale, but now the numerical label isn't what I want:
Have you tried horizon.extent? It lets you specify the [min, max] value for the horizon chart. By default, a linear scale will be created to map values within the extent to the pixels within the chart's height (specified with `horizon.height or default to 30 pixels).

Matplotlib/Pyplot: How to zoom subplots together?

I have plots of 3-axis accelerometer time-series data (t,x,y,z) in separate subplots I'd like to zoom together. That is, when I use the "Zoom to Rectangle" tool on one plot, when I release the mouse all 3 plots zoom together.
Previously, I simply plotted all 3 axes on a single plot using different colors. But this is useful only with small amounts of data: I have over 2 million data points, so the last axis plotted obscures the other two. Hence the need for separate subplots.
I know I can capture matplotlib/pyplot mouse events (http://matplotlib.sourceforge.net/users/event_handling.html), and I know I can catch other events (http://matplotlib.sourceforge.net/api/backend_bases_api.html#matplotlib.backend_bases.ResizeEvent), but I don't know how to tell what zoom has been requested on any one subplot, and how to replicate it on the other two subplots.
I suspect I have the all the pieces, and need only that one last precious clue...
-BobC
The easiest way to do this is by using the sharex and/or sharey keywords when creating the axes:
from matplotlib import pyplot as plt
ax1 = plt.subplot(2,1,1)
ax1.plot(...)
ax2 = plt.subplot(2,1,2, sharex=ax1)
ax2.plot(...)
You can also do this with plt.subplots, if that's your style.
fig, ax = plt.subplots(3, 1, sharex=True, sharey=True)
Interactively this works on separate axes
for ax in fig.axes:
ax.set_xlim(0, 50)
fig.draw()