Connect with us

TOP SCEINCE

How do you make a robot smarter? Program it to know what it doesn’t know

Published

on

How do you make a robot smarter? Program it to know what it doesn’t know


Modern robots know how to sense their environment and respond to language, but what they don’t know is often more important than what they do know. Teaching robots to ask for help is key to making them safer and more efficient.

Engineers at Princeton University and Google have come up with a new way to teach robots to know when they don’t know. The technique involves quantifying the fuzziness of human language and using that measurement to tell robots when to ask for further directions. Telling a robot to pick up a bowl from a table with only one bowl is fairly clear. But telling a robot to pick up a bowl when there are five bowls on the table generates a much higher degree of uncertainty — and triggers the robot to ask for clarification.

Because tasks are typically more complex than a simple “pick up a bowl” command, the engineers use large language models (LLMs) — the technology behind tools such as ChatGPT — to gauge uncertainty in complex environments. LLMs are bringing robots powerful capabilities to follow human language, but LLM outputs are still frequently unreliable, said Anirudha Majumdar, an assistant professor of mechanical and aerospace engineering at Princeton and the senior author of a study outlining the new method.

“Blindly following plans generated by an LLM could cause robots to act in an unsafe or untrustworthy manner, and so we need our LLM-based robots to know when they don’t know,” said Majumdar.

The system also allows a robot’s user to set a target degree of success, which is tied to a particular uncertainty threshold that will lead a robot to ask for help. For example, a user would set a surgical robot to have a much lower error tolerance than a robot that’s cleaning up a living room.

“We want the robot to ask for enough help such that we reach the level of success that the user wants. But meanwhile, we want to minimize the overall amount of help that the robot needs,” said Allen Ren, a graduate student in mechanical and aerospace engineering at Princeton and the study’s lead author. Ren received a best student paper award for his Nov. 8 presentation at the Conference on Robot Learning in Atlanta. The new method produces high accuracy while reducing the amount of help required by a robot compared to other methods of tackling this issue.

The researchers tested their method on a simulated robotic arm and on two types of robots at Google facilities in New York City and Mountain View, California, where Ren was working as a student research intern. One set of hardware experiments used a tabletop robotic arm tasked with sorting a set of toy food items into two different categories; a setup with a left and right arm added an additional layer of ambiguity.

The most complex experiments involved a robotic arm mounted on a wheeled platform and placed in an office kitchen with a microwave and a set of recycling, compost and trash bins. In one example, a human asks the robot to “place the bowl in the microwave,” but there are two bowls on the counter — a metal one and a plastic one.

The robot’s LLM-based planner generates four possible actions to carry out based on this instruction, like multiple-choice answers, and each option is assigned a probability. Using a statistical approach called conformal prediction and a user-specified guaranteed success rate, the researchers designed their algorithm to trigger a request for human help when the options meet a certain probability threshold. In this case, the top two options — place the plastic bowl in the microwave or place the metal bowl in the microwave — meet this threshold, and the robot asks the human which bowl to place in the microwave.

In another example, a person tells the robot, “There is an apple and a dirty sponge … It is rotten. Can you dispose of it?” This does not trigger a question from the robot, since the action “put the apple in the compost” has a sufficiently higher probability of being correct than any other option.

“Using the technique of conformal prediction, which quantifies the language model’s uncertainty in a more rigorous way than prior methods, allows us to get to a higher level of success” while minimizing the frequency of triggering help, said the study’s senior author Anirudha Majumdar, an assistant professor of mechanical and aerospace engineering at Princeton.

Robots’ physical limitations often give designers insights not readily available from abstract systems. Large language models “might talk their way out of a conversation, but they can’t skip gravity,” said coauthor Andy Zeng, a research scientist at Google DeepMind. “I’m always keen on seeing what we can do on robots first, because it often sheds light on the core challenges behind building generally intelligent machines.”

Ren and Majumdar began collaborating with Zeng after he gave a talk as part of the Princeton Robotics Seminar series, said Majumdar. Zeng, who earned a computer science Ph.D. from Princeton in 2019, outlined Google’s efforts in using LLMs for robotics, and brought up some open challenges. Ren’s enthusiasm for the problem of calibrating the level of help a robot should ask for led to his internship and the creation of the new method.

“We enjoyed being able to leverage the scale that Google has” in terms of access to large language models and different hardware platforms, said Majumdar.

Ren is now extending this work to problems of active perception for robots: For instance, a robot may need to use predictions to determine the location of a television, table or chair within a house, when the robot itself is in a different part of the house. This requires a planner based on a model that combines vision and language information, bringing up a new set of challenges in estimating uncertainty and determining when to trigger help, said Ren.



Source link

Continue Reading
Click to comment

Leave a Reply

TOP SCEINCE

New understanding of fly behavior has potential application in robotics, public safety

Published

on

By

How do you make a robot smarter? Program it to know what it doesn’t know


Why do flies buzz around in circles when the air is still? And why does it matter?

In a paper published online July 26, 2024 by the scientific journal Current Biology, University of Nevada, Reno Assistant Professor Floris van Breugel and Postdoctoral Researcher S. David Stupski respond to this up-until-now unanswered question. And that answer could hold a key to public safety — specifically, how to better train robotic systems to track chemical leaks.

“We don’t currently have robotic systems to track odor or chemical plumes,” van Breugel said. “We don’t know how to efficiently find the source of a wind-borne chemical. But insects are remarkably good at tracking chemical plumes, and if we really understood how they do it, maybe we could train inexpensive drones to use a similar process to find the source of chemicals and chemical leaks.”

A fundamental challenge in understanding how insects track chemical plumes — basically, how does the fly find the banana in your kitchen? — is that wind and odors can’t be independently manipulated.

To address this challenge, van Breugel and Stupski used a new approach that makes it possible to remotely control neurons — specifically the “smell” neurons — on the antennae of flying fruit flies by genetically introducing light-sensitive proteins, an approach called optogenetics. These experiments, part of a $450,000 project funded through the Air Force Office of Scientific Research, made it possible to give flies identical virtual smell experiences in different wind conditions.

What van Breugel and Stupski wanted to know: how do flies find an odor when there’s no wind to carry it? This is, after all, likely the wind experience of a fly looking for a banana in your kitchen. The answer is in the Current Biology article, “Wind Gates Olfaction Driven Search States in Free Flight.” The print version will appear in the Sept. 9 issue.

Flies use environmental cues to detect and respond to air currents and wind direction to find their food sources, according to van Breugel. In the presence of wind, those cues trigger an automatic “cast and surge” behavior, in which the fly surges into the wind after encountering a chemical plume (indicating food) and then casts — moves side to side — when it loses the scent. Cast-and-surge behavior long has been understood by scientists but, according to van Breugel, it was fundamentally unknown how insects searched for a scent in still air.

Through their work, van Breugel and Stupski uncovered another automatic behavior, sink and circle, which involves lowering altitude and repetitive, rapid turns in a consistent direction. Flies perform this innate movement consistently and repetitively, even more so than cast-and-surge behavior.

According to van Breugel, the most exciting aspect of this discovery is that it shows flying flies are clearly able to assess the conditions of the wind — its presence, and direction — before deploying a strategy that works well under these conditions. The fact that they can do this is actually quite surprising — can you tell if there is a gentle breeze if you stick your head out of the window of a moving car? Flies aren’t just reacting to an odor with the same preprogrammed response every time like a simple robot, they are responding in context-appropriate manner. This knowledge potentially could be applied to train more sophisticated algorithms for scent-detecting drones to find the source of chemical leaks.

So, the next time you try to swat a fly in your home, consider the fact that flies might actually be a little more aware of some of their natural surroundings than you are. And maybe just open a window to let it out.



Source link

Continue Reading

TOP SCEINCE

New drug shows promise in clearing HIV from brain

Published

on

By

How do you make a robot smarter? Program it to know what it doesn’t know


An experimental drug originally developed to treat cancer may help clear HIV from infected cells in the brain, according to a new Tulane University study.

For the first time, researchers at Tulane National Primate Research Center found that a cancer drug significantly reduced levels of SIV, the nonhuman primate equivalent of HIV, in the brain by targeting and depleting certain immune cells that harbor the virus.

Published in the journal Brain, this discovery marks a significant step toward eliminating HIV from hard-to-reach reservoirs where the virus evades otherwise effective treatment.

“This research is an important step in tackling brain-related issues caused by HIV, which still affect people even when they are on effective HIV medication,” said lead study author Woong-Ki Kim, PhD, associate director for research at Tulane National Primate Research Center. “By specifically targeting the infected cells in the brain, we may be able to clear the virus from these hidden areas, which has been a major challenge in HIV treatment.”

Antiretroviral therapy (ART) is an essential component of successful HIV treatment, maintaining the virus at undetectable levels in the blood and transforming HIV from a terminal illness into a manageable condition. However, ART does not completely eradicate HIV, necessitating lifelong treatment. The virus persists in “viral reservoirs” in the brain, liver, and lymph nodes, where it remains out of reach of ART.

The brain has been a particularly challenging area for treatment due to the blood-brain barrier — a protective membrane that shields it from harmful substances but also blocks treatments, allowing the virus to persist. In addition, cells in the brain known as macrophages are extremely long-lived, making them difficult to eradicate once they become infected.

Infection of macrophages is thought to contribute to neurocognitive dysfunction, experienced by nearly half of those living with HIV. Eradicating the virus from the brain is critical for comprehensive HIV treatment and could significantly improve the quality of life for those with HIV-related neurocognitive problems.

Researchers focused on macrophages, a type of white blood cell that harbors HIV in the brain. By using a small molecule inhibitor to block a receptor that increases in HIV-infected macrophages, the team successfully reduced the viral load in the brain. This approach essentially cleared the virus from brain tissue, providing a potential new treatment avenue for HIV.

The small molecule inhibitor used, BLZ945, has previously been studied for therapeutic use in amyotrophic lateral sclerosis (ALS) and brain cancer, but never before in the context of clearing HIV from the brain.

The study, which took place at the Tulane National Primate Research Center, utilized three groups to model human HIV infection and treatment: an untreated control group, and two groups treated with either a low or high dose of the small molecule inhibitor for 30 days. The high-dose treatment lead to a notable reduction in cells expressing HIV receptor sites, as well as a 95-99% decrease in viral DNA loads in the brain .

In addition to reducing viral loads, the treatment did not significantly impact microglia, the brain’s resident immune cells, which are essential for maintaining a healthy neuroimmune environment. It also did not show signs of liver toxicity at the doses tested.

The next step for the research team is to test this therapy in conjunction with ART to assess its efficacy in a combined treatment approach. This could pave the way for more comprehensive strategies to eradicate HIV from the body entirely.

This research was funded by the National Institutes of Health, including grants from the National Institute of Mental Health and the National Institute of Neurological Disorders and Stroke, and was supported with resources from the Tulane National Primate Research Center base grant of the National Institutes of Health, P51 OD011104.



Source link

Continue Reading

TOP SCEINCE

Chemical analyses find hidden elements from renaissance astronomer Tycho Brahe’s alchemy laboratory

Published

on

By

How do you make a robot smarter? Program it to know what it doesn’t know


In the Middle Ages, alchemists were notoriously secretive and didn’t share their knowledge with others. Danish Tycho Brahe was no exception. Consequently, we don’t know precisely what he did in the alchemical laboratory located beneath his combined residence and observatory, Uraniborg, on the now Swedish island of Ven.

Only a few of his alchemical recipes have survived, and today, there are very few remnants of his laboratory. Uraniborg was demolished after his death in 1601, and the building materials were scattered for reuse.

However, during an excavation in 1988-1990, some pottery and glass shards were found in Uraniborg’s old garden. These shards were believed to originate from the basement’s alchemical laboratory. Five of these shards — four glass and one ceramic — have now undergone chemical analyses to determine which elements the original glass and ceramic containers came into contact with.

The chemical analyses were conducted by Professor Emeritus and expert in archaeometry, Kaare Lund Rasmussen from the Department of Physics, Chemistry, and Pharmacy, University of Southern Denmark. Senior researcher and museum curator Poul Grinder-Hansen from the National Museum of Denmark oversaw the insertion of the analyses into historical context.

Enriched levels of trace elements were found on four of them, while one glass shard showed no specific enrichments. The study has been published in the journal Heritage Science.

“Most intriguing are the elements found in higher concentrations than expected — indicating enrichment and providing insight into the substances used in Tycho Brahe’s alchemical laboratory,” said Kaare Lund Rasmussen.

The enriched elements are nickel, copper, zinc, tin, antimony, tungsten, gold, mercury, and lead, and they have been found on either the inside or outside of the shards.

Most of them are not surprising for an alchemist’s laboratory. Gold and mercury were — at least among the upper echelons of society — commonly known and used against a wide range of diseases.

“But tungsten is very mysterious. Tungsten had not even been described at that time, so what should we infer from its presence on a shard from Tycho Brahe’s alchemy workshop?,” said Kaare Lund Rasmussen.

Tungsten was first described and produced in pure form more than 180 years later by the Swedish chemist Carl Wilhelm Scheele. Tungsten occurs naturally in certain minerals, and perhaps the element found its way to Tycho Brahe’s laboratory through one of these minerals. In the laboratory, the mineral might have undergone some processing that separated the tungsten, without Tycho Brahe ever realizing it.

However, there is also another possibility that Professor Kaare Lund Rasmussen emphasizes has no evidence whatsoever — but which could be plausible.

Already in the first half of the 1500s, the German mineralogist Georgius Agricola described something strange in tin ore from Saxony, which caused problems when he tried to smelt tin. Agricola called this strange substance in the tin ore “Wolfram” (German for Wolf’s froth, later renamed to tungsten in English).

“Maybe Tycho Brahe had heard about this and thus knew of tungsten’s existence. But this is not something we know or can say based on the analyses I have done. It is merely a possible theoretical explanation for why we find tungsten in the samples,” said Kaare Lund Rasmussen.

Tycho Brahe belonged to the branch of alchemists who, inspired by the German physician Paracelsus, tried to develop medicine for various diseases of the time: plague, syphilis, leprosy, fever, stomach aches, etc. But he distanced himself from the branch that tried to create gold from less valuable minerals and metals.

In line with the other medical alchemists of the time, he kept his recipes close to his chest and shared them only with a few selected individuals, such as his patron, Emperor Rudolph II, who allegedly received Tycho Brahe’s prescriptions for plague medicine.

We know that Tycho Brahe’s plague medicine was complicated to produce. It contained theriac, which was one of the standard remedies for almost everything at the time and could have up to 60 ingredients, including snake flesh and opium. It also contained copper or iron vitriol (sulphates), various oils, and herbs.

After various filtrations and distillations, the first of Brahe’s three recipes against plague was obtained. This could be made even more potent by adding tinctures of, for example, coral, sapphires, hyacinths, or potable gold.

“It may seem strange that Tycho Brahe was involved in both astronomy and alchemy, but when one understands his worldview, it makes sense. He believed that there were obvious connections between the heavenly bodies, earthly substances, and the body’s organs. Thus, the Sun, gold, and the heart were connected, and the same applied to the Moon, silver, and the brain; Jupiter, tin, and the liver; Venus, copper, and the kidneys; Saturn, lead, and the spleen; Mars, iron, and the gallbladder; and Mercury, mercury, and the lungs. Minerals and gemstones could also be linked to this system, so emeralds, for example, belonged to Mercury,” explained Poul Grinder-Hansen.

Kaare Lund Rasmussen has previously analyzed hair and bones from Tycho Brahe and found, among other elements, gold. This could indicate that Tycho Brahe himself had taken medicine that contained potable gold.



Source link

Continue Reading

Trending