Sam Cobb's Amazing Automata Sculptures! (6 min 25 sec)

A rare glimpse of how a single motor can control multiple limbs via a cam shaft. That's about all of the inner workings of an automaton you'll find on the goog. Making a self propelled quadruped by having a single powerful motor turn all the limbs was appealing but much more complex than multiple motors. Then, fully exploiting the ability of legs to overcome obstacles would require a full elevation map of the terrain & the logic to avoid the obstacles. It would be as complicated as building a Falcon 9, taking multiple lifetimes.


The good news is the DJI Mavic appears to be the 1st quad copter suitable for replacing ground based exercise robots. It would only work for 1 mile at a time, but it's finally stable enough to be controlled by the same single handed controller that the ground based robots use.

Blackhawk run timelapse in 4k (4 min 50 sec)

Pleasanton run in 4k (3 min 46 sec)

...Continue Reading

Managed to reconfigure it so the camera mounts the right way. That got it an inch higher & eliminated the need to rotate the video. Requiring any kind of processing just to preview the video was a disaster.

Optimized the UARTs so the feedback now goes at 1960Hz instead of 1830Hz. The trick was not doing anything else until the packet was transmitted. More speed came from setting the IMU sample rate divider to 0. It doesn't cause aliasing. It just reads the sample with lower latency. This might have reduced the bobbing slightly.

Another attempt to handle the UART transmitter using interrupts failed. It didn't matter since context switching would count as doing something else while the packet was transmitted. Tried delaying the feedback commands by 1 packet so the next IMU reading could be read before the previous readout's feedback calculations were done. While it would have increased the rate, it would have added latency. The increased latency was a failure.

Finally, extended the motor mixing to allow it to be used for selfies & near the ground. This took a bit of geometry.

Discovered the P & D feedback terms in their only working configuration don't fight each other. D actually adds to P, so fighting terms aren't the reason it undershoots. Tried maximizing P & D with no obvious improvement. Did confirm it slightly undershoots the feedback, leading to bobbing when running.

Tried replacing the traditional IMU blending algorithm, which...Continue Reading

The last 18.6 mile drive was a long long time ago:

https://www.rcgroups.com/forums/show....php?t=2742360

For today's 18.5 mile drive, put battery 1 on the bottom & battery 2 on top. Battery 1 still gave 3.4Ah & battery 2 gave 4.5Ah, so the cable was not the limiting factor & hauling around drinks didn't make any difference. It was the internal resistance of the batteries. Battery 2 was dead by the end.


In other news, a rare teardown of the DJI Mavic rounded the blogs.

DJI Mavic detailed Teardown (19 min 22 sec)

It was another frustrating experience watching a millenial struggle through simple concepts & have no idea how things worked the way previous generations did.


The mane nugget was how the IMU has evolved from being randomly scattered on the mane board to being encapsulated in a very large, weighted, shock dampered enclosure, connected by a very long ribbon cable to reduce coupling. It matches what lions discovered years ago.


It was enlightening to discover lions were on the right track in their treatment of the IMU with a lot more packaging than the main processor, while everyone else was putting everything on a single board.

The answer is yes. Cinelerra can play 3840x2160 in a 2560x1600 window at 30fps without any hardware acceleration, on the 7 year old AMD. For all those years, it seemed hopeless, but it had the power long before 4k cameras were around. It doesn't work so well in a 3840x2160 window, but the lion kingdom can't afford a 4k monitor. It requires some diabolical hacks, without which it only did 15fps for all those years.


XMovie used a file -> BGR8888 bitmap -> window pipeline back when computers could barely play 720x480. The only way to watch a movie was on XMovie. When computers got fast enough, XMovie went away & Cinelerra got away with using file -> YUV888 intermediate -> BGR8888 bitmap -> window to play everything up to 1920x1080. Once again, the limit of computing power has been reached & the intermediate copy needs to be skipped again.


It was surprising that file -> YUV888 intermediate -> OpenGL was slower than file -> BGR8888 bitmap -> window. OpenGL was always slower than software on this system, but it used a 10 year old graphics card.


The idea of optimizing it occurred while adding a colorspace converter for the new HDR codecs & remembering all the codecs still contain the scaling routines for XMovie to play on low end machines 17 years ago. There's also the fact that virtual machines, faster CPUs, & vanishing interest in PC video has made hardware acceleration in Linux go away & the old software BGR8888 bitmap has...Continue Reading

Amusing that after all the money invested in the Antares rocket, having it explode then redesigning it for another engine, it didn't have enough power so they ended up going back to the Atlas V. Atlas V allows 700 more lbs of payload to the space station, bringing it to 8157 lbs. The most the Falcon 9 moved to the space station was 6914 lbs, but it recovered its 1st stage. The Falcon 9 is gone until next year, which means all the money spent on the commercial cargo program led to just 1 new spaceship launching on the same old rocket.


https://www.nasaspaceflight.com/2016...e-antares-rtf/

Today's supermoon vs 2014 eclipse. It was only 1 pixel bigger than the 2014 supermoon.

During a 10 mile drive,

Berkeley run 2 (3 min 2 sec)

passed the robotics meetup.

https://twitter.com/carlbass/status/797908532360134656

It was intriguing to see another lunchbox in his photos.

The warehouse had been recently renovated. There was a bathroom & desperately needed drinks for the lions running 10 miles with robots. The CEO of Autodesk owned it & might be trying to start a self driving car startup. Of course, all the bigwigs were long gone by the time the lion kingdom drove up & the crowd which much thinner.

There were a few hobbyists working on similar 1/10 vehicles. They were all using webcams & pi cams to try to navigate the indoor track. Documented just the track, in case inspiration struck. It's basically a line following problem, with lines of much less contrast than historic contests.

Started out in robotics by reading http://www.robotroom.com/ That guy was incredibly dedicated to just line following, for many years. Line following remanes a toy problem. The lion kingdom would be ecstatic if he focused just on a segment of pavement with variable shadows, rather than getting distracted by line following.

Speaking of Autodesk, as of 2013 SpaceX was using Siemens NX, not Autodesk Inventor for its CAD. Autodesk needs to buy out Siemens.

Aquatic pier run timelapse 20:1 (4 min 31 sec)

A 3rd attempt to make this one look good. Software stabilization was horrible for most of it. Raw footage only looked stable when accelerated 20:1, but not 10:1 because the gimbal has an oscillating movement. So made the salient feature a software stabilized 5:1 segment with the rest at 20:1.

Then there was the new & improved bridge timelapse with no software stabilization:

Bay bridge bike path (6 min 27 sec)

...Continue Reading

So fired up the video editor & made a spat of brushless gimbal timelapses for the 1st time in 3 years.

Pier 39 run timelapse (5 min 47 sec)

Coit timelapse (5 min 5 sec)

...Continue Reading

So the helium bottles on the Falcon 9 are pressurized to 5000psi. A household propane tank is 300psi at most. They're submerged in the LOX to increase the amount of helium they can store & save space. In previous rockets, they sit in the LOX bath long after filling is complete, to equalize the temperature before being pressurized.

The new Falcon 9 has to take off immediately after tanking, in order to keep the LOX supercooled far below its boiling point. The helium bottles have to be filled before the LOX filling is complete & before their temperature equalizes. The trick is getting the helium the right temperature so the interior & exterior of the tank are the same temperature even as the exterior is still cooling down. There's probably a temperature differential they have to be within. If the helium is too cold, the exterior of the tank is shrinking slower than the interior & it sheers. If it's too hot, the interior is shrinking slower than the exterior & it sheers. An easy solution would be to move the helium tanks to the bottom of the LOX tank so they have more time to cool, but this would add weight & SpaceX wants to save money by tweeking the filling procedures.

Now we know the latest explosion happened when the helium bottles were being filled & someone in LA knew immediately what happened. Still not clear how a bursting carbon fiber tank in a pure LOX environment can ignite. Perhaps it damaged something on the service tower which ignited it.

It helps to remember the 1st walking animals were 6 legged, 8 legged, or hundreds of legged. It took millions of years of evolution to go from flapping our entire bodies to legs.

Bio-Inspired Recon Robot (Man-Packable, All-Terrain Situational Awareness) (4 min 52 sec)

This one slowly but surely does the business. It took a few million more years to gain enough intelligence to get by on 4 legs.

A Simple 4-Legged Walking Robot (0 min 9 sec)

...Continue Reading