How to spoof robotics demonstration to make money and have fun

How to spoof robotics demonstration to make money and have fun: A roboticist dressed for the cold gave Big Dog a vigorous kick for the camera in March 2008. The DARPA-funded robot began to buzz, then faltered in the icy parking lot before catching its balance again. The top comment on the video begs, “PLEASE DO NOT KICK THE WALKING PROTOTYPE DEATH MECH.” “IT WILL CALL TO ME REMINDER.”

Someone another says, “Creepy as hell.” “Considerate the possibility that you saw that thing approaching you one day while out for a walk in the woods.” In those pre-Black Mirror days, social media accounts and gadget blogs would alternately use terms like “robopocalypse” and “terrifying,” before the show provided an even more blunt summary. Boston Dynamics saw success. There are now 17 million views on the video. It was the first in a long line of viral hits that still happen today.

The impact that such virality has on Boston Dynamics’ subsequent growth into one of the most recognisable robotics businesses in the world cannot be overstated. Big Dog and its offspring, such as Spot and Atlas, have been praised, ridiculed, mocked, and even featured in a beer advertisement for Sam Adams. In addition to creating some of the most sophisticated mechatronics in the world, the Boston Dynamics team has shown itself to be a very astute marketeer.
It is important to acknowledge the impact that these films have had in promoting robotics.

It is possible that innumerable roboticists who are presently flourishing in the area were inspired by films similar to this one. Numerous other firms have used this concept with varying degrees of success. Without a doubt, none of those businesses for which Boston Dynamics may have taken some short cuts can be held accountable.

Viral robot videos have evolved over the last few decades from technologicalrati curiosity items to attention-grabbing successes that are discovered via TikTok and YouTube. The incentive to smooth the edges has grown along with the possible benefits. The status of CGI, which many viewers no longer can tell apart from reality, further complicates things. Our propensity to trust bogus news and videos is mostly influenced by confirmation bias, novelty appeal, and a lack of technical knowledge.

For example, you may overlook the fact that the typical TikTok viewer is unaware of the complexities involved with generalisation. It is possible that many roboticists accidentally fueled the flames by suggesting that the systems we see in films are “general purpose.” Though multipurpose, we are still a long way from having robots capable of any work without being constrained by technology.

The videos you see are often the result of months or years of labour. The hours of footage of a robot fumbling, sputtering, or halting short that did not make the final edit are stored somewhere on a hard drive. For this very reason, I have urged businesses to show some of these movies to the TechCrunch readership. Not surprisingly, not many people have accepted my offer. I think a lot of this has to do with how individuals interpret this kind of information. The many hours and days of trial and error in robotics are a testament to the amount of effort put in to produce the finished result. However, such robot failures can be seen by the general public as a failure on the part of the roboticists themselves.

I gave Boston Dynamics credit for the “blooper reel” they released, which showed Atlas losing its balance and toppling in between impressive parkour manoeuvres, back in a 2023 edition of Actuator (RIP). As usual, much more material was cut than made it to the final edit. It is the way things are, even in the absence of robots.

A few weeks ago, after seeing her fantastic newish film, “Showing Up,” I went to a discussion by filmmaker Kelly Reichardt. She restated that old chestnut of W.C. Fields’s about never dealing with kids or animals. Generally speaking, sophisticated mechatronics would be a good addition to that list.

Creative editing is simply one of many possible techniques to enhance a robotics demo, in addition to CG and renderings. Most of the time, there is no malevolent intent. Musicians often express to me on my show that they feel powerless over a song after it is placed into the public domain. I think the same can be said for video to some level. Decisions are taken to simplify and enhance the display. When creating engaging internet videos, they are a must. But context is the first casualty, especially in the TikTok era.

The information that should be included in a robotics demo is not prescribed. But the more I consider it, the more I think there need to be some clear rules, at the very least. I do not practise robotics. With a BA in creative writing, I am simply a geek. I do, however, often discuss the topic with individuals considerably wiser than I am.

Prior to CES, I saw a post on LinkedIn that drew the attention of many people in the robotics field. Written by Brad Porter, the CEO and creator of Collaborative robots, who oversaw Amazon’s industrial robots division before, it was published. Although I do not often suggest LinkedIn followers, he is an excellent one if you are interested in the field.
Porter writes in the article that while there were not many impressive robotics demonstrations at CES (which it wasn’t), “there are also a lot of amazing trick-shot videos out there.” It is difficult to distinguish stagecraft from reality. The CEO was not making any allusions to the bad connotations that a term like “stagecraft” may have in this situation. Rather, he was only advising people to examine these films with a critical and maybe even sceptical eye.

I have been covering this area for a while, so I have honed some abilities in identifying robotic kayfabe. But when something seems odd with a demo, I still often turn to industry veterans like Porter. Naturally, not every viewer has the same access to these people or my experience. On the other hand, they may arm themselves by understanding how these movies are sweetened, whether intentionally or not.

Porter makes five distinct observations. Firstly, there is “stop-motion.” This is a reference to a series of quick adjustments that give the impression that the robot is moving in ways that it is not really able to.

He advises viewers to “be wary” of robotics videos that include a lot of frame skips or camera cuts. “It is impressive that Boston Dynamics frequently uses a single cut without any camera cuts in their videos.”

Second is modelling. In actuality, this is the CG example I mentioned before. In robotic deployment, simulation has emerged as a key component. It enables users to quickly simulate thousands of scenarios at once. In recent years, robotic simulation, like other computer graphics, has become more lifelike. It is not difficult to create and distribute a realistic simulation by oneself. Instead, the problem comes from passing off such falsehoods as true.

Third issue has a catchy moniker. The reason Wizard of Oz demonstrations are named that is because [person] behind the curtain does a lot of hard lifting (do not pay attention). Porter uses the Mobile ALOHA demo from Stanford as an example. I firmly feel that the choice to use off-screen teleop for the (still really spectacular) demo was made without malice. Tony Zhao, the “robot operator,” really makes an appearance in both the end credits and the video.
“A few more inches to the left in the framing of the Optimus laundry video, and you would have missed what appears to be a tele-op hand controlling Tesla Bot,” I said at the time. Of course, there is nothing wrong with tele-ops. It is great for teaching, troubleshooting, and doing very specialised activities like surgery, among other things. However, knowing what we are (and are not) seeing is wonderful. I think it is clear that the original poster left out important details because he knew his audiences and admirers would fill in the blanks with their own interpretations of what they saw depending on how they felt about the message.

To say that Musk is purposefully obfuscating the facts in this instance may not be accurate. “Important note: Optimus cannot yet do this autonomously, but certainly will be able to do this fully autonomously and in an arbitrary environment (will not require a fixed table with box that has only one shirt)” he added twenty-three minutes after the first tweet.

The famous observation made by Mark Twain was that “a lie can travel halfway around the world while the truth is still putting on its shoes.” You may use a similar technique to streaming video. Of course, the first tweet is not technically false, but it is undoubtedly an omission. It is the venerable journalistic practice of concealing your edits on page A12. The original mistake will be seen by many more people.

Once again, I am not here to judge whether or not the first omission was deliberate (if you want to give yourself a break in this instance, you may definitely see the subsequent tweet as a sincere attempt to clarify incomplete context). In this particular case, I anticipate that the majority of viewpoints will be closely associated with an individual’s subjective emotions towards the writer.

The second example given by Porter is “Single-task Reinforcement Learning.” You may read more about reinforcement learning here, but to keep things simple in an essay that is not even close to that length, let us just state that it is a method of teaching robots to carry out tasks that need repeated real-world trial and error.

Porter writes, “Open a door, stack a block, turn a crank.” “It is impressive to learn these tasks; they look impressive and are impressive.” However, this can be worked out in a few months by a skilled RL engineer. To making it resilient to several small modifications is a step more difficult. However, generalising to other related activities is rather challenging. Look for many training tasks to see if it can generalise.

Like teleop, reinforcement learning has no drawbacks at all. These are both very useful resources for robot operation and training. All you have to do is make them known as plainly as you can.

Porter’s last piece of advice is to keep an eye out for any omissions and the surroundings. He brings up the then-current footage of Figure’s coffee-making humanoid. He describes it as “fluid, single-cut, shows robustness to failure modes.” There is no sign of robotics having its ChatGPT moment here—it is still simply a single task. The quality of the production is excellent. However, you will see that the robot can only carry a Keurig cup and nothing heavier. Although it has been done, they do not demonstrate that. Is it possible that the robot lacks such strength?

Porter was eager to emphasise that these criticisms do not take away from what is really amazing technology when I chatted with him about the nuances of the article today. But the problem is that our minds prefer to fill in the blanks. It is a common misconception that robots are human like us and can learn in the same manner as humans. However, it is important to remember that just because a robot can open one door does not mean it can open another or even the same door in a different light. Our preconceived notions of what robots may and cannot do in 2024 have also been shaped by TV shows and motion pictures.

Finally, speed is a point that was left out of the article. Because technology may be excruciatingly sluggish at times, it is customary to accelerate processes. Universities and other research centres often do an excellent job of using a text overlay to highlight this. This is the appropriate method. Put the important details on display in a manner that makes it hard for a click-happy influencer to miss. In fact, 1X received its name because of this phenomena.
This fact is highlighted in a new video from the firm that shows off how it uses neural networks. The business states, “There is no teleoperation, computer graphics, cutting, video speedups, or scripted trajectory playback in this video.” All of it is managed by neural networks. The result is a three-minute film that, in comparison to previous humanoid displays, might seem excruciatingly sluggish.

Similar to the blooper videos, I support any and all forms of openness. There is nothing wrong with increasing speed for robots that move really slowly as long as you follow these three important guidelines:

Open up
Open up
Open up

Similar to the songwriter, corporations must accept that once a video is in the public domain, it is out of their control. However, consider this: Did I take all reasonable steps to stop the spread of such forgeries?

Hopefully, films of this kind will not be subject to the same truth-in-advertising laws that apply to television commercials. But I would really want to see a consortium of roboticists get together to standardise the ways in which these disclosures can — and ought to — operate.