Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI

A new artificial muscle could let humanoid robots lift 4,400 times their weight

A new material bends that rule.

Researchers in South Korea say they have built a soft, magnetic artificial muscle that hits hard numbers without turning into a stiff piston. The material flexes, contracts and relaxes like flesh, yet ramps up stiffness on demand when asked to do real work. That mix has long sat out of reach for humanoid robots that need both agility and strength.

Most humanoids move with a cocktail of motors, gears and pneumatic lines. These systems deliver power, but they also add bulk and make contact risky. Soft actuators change the equation. They integrate into limbs, cushion impacts and tolerate misalignment. They also weigh far less than hydraulic stacks and slot neatly inside compact forms like hands, faces and torsos.

The cost of thinking: Reasoning models share aspects of information processing with human brains

Large language models (LLMs) like ChatGPT can write an essay or plan a menu almost instantly. But until recently, it was also easy to stump them. The models, which rely on language patterns to respond to users’ queries, often failed at math problems and were not good at complex reasoning. Suddenly, however, they’ve gotten a lot better at these things.

A new generation of LLMs known as reasoning models are being trained to solve complex problems. Like humans, they need some time to think through problems like these—and remarkably, scientists at MIT’s McGovern Institute for Brain Research have found that the kinds of problems that require the most processing from reasoning models are the very same problems that people need to take their time with.

In other words, they report in the journal PNAS, the “cost of thinking” for a reasoning model is similar to the cost of thinking for a human.

AI-Designed Proteins Can Boost Production of T Cells

Daley and Mout added that the team is excited that the approach can guide T cells to tumors, stimulate their cancer cell-killing abilities, and overcome immune suppression by the tumor microenvironment.

Mout, who trained in the lab of Nobel Prize-winning co-author and Rosetta creator David Baker, is especially enthusiastic about the technology’s far-reaching potential.

“Our goal is to develop next-generation immunotherapies and cancer vaccines,” he said.

Wireless image transmission technique filters redundant data intuitively—just like a human

A new AI-driven technology developed by researchers at UNIST promises to significantly reduce data transmission loads during image transfer, paving the way for advancements in autonomous vehicles, remote surgery and diagnostics, and real-time metaverse rendering—applications that demand rapid, large-scale visual data exchange without delay.

Led by Professor Sung Whan Yoon from the Graduate School of Artificial Intelligence at UNIST, the research team developed Task-Adaptive Semantic Communication, an innovative wireless image transmission method that selectively transmits only the most essential semantic information relevant to the specific task. Their study is published in the IEEE Journal on Selected Areas in Communications.

Current wireless image transmission methods compress entire images without considering their underlying semantic structures—such as objects, layout, and relationships—resulting in bandwidth limitations and transmission delays that hinder high-resolution image sharing.

Immune checkpoint inhibitor-associated myocarditis and pericarditis: a pharmacovigilance study based on the FAERS database

Immune checkpoint inhibitors (ICIs) are medications used in cancer immunotherapy. However, treatment with ICIs may lead to adverse effects, particularly myocarditis and pericarditis. This practical pharmacovigilance study investigates the relationship between ICIs and myocarditis and pericarditis using the FAERS (U.S. FDA Adverse Event Reporting System) database.

Data on myocarditis and pericarditis related to ICIs were extracted from the FAERS database for the period from 2014Q1 to 2023Q4. Data mining was performed using the Bayesian Confidence Propagation Neural Network (BCPNN) and the Reporting Odds Ratio (ROR).

A total of 1,112 cases involving 1,134 adverse event (AE) reports related to ICIs-associated noninfectious myocarditis/pericarditis (NM/P) were extracted from the FAERS database. After excluding reports with missing data, the primary reporters were physicians, consumers, and pharmacists, with the United States and Japan being the main reporting countries. The cases showed a greater percentage of males than females, with a median age of 67 years, a median weight of 65 kg, and a median onset time of 28 days. The signal strength of ICIs-associated NM/P, from highest to lowest, was as follows: Pembrolizumab (ROR: 12.32, 95% CI: 11.28–13.45, IC 025: 3.45) Nivolumab (ROR: 11.23, 95% CI: 10.13–12.44, IC 025: 3.30) Atezolizumab (ROR: 10.62, 95% CI: 8.67–13.02, IC 025: 3.10) Ipilimumab (ROR: 10.25, 95% CI: 8.34–12.58, IC 025: 3.04) Durvalumab (ROR: 9.25, 95% CI: 7.21–11.88, IC 025: 2.83).

Agility’s ‘hardest working’ humanoid robot hits 100,000-tote milestone

Oregon-based robotics company Agility Robotics announced Thursday that its humanoid robot Digit has moved more than 100,000 totes at a GXO Logistics facility in Flowery Branch, Georgia.

This milestone marks a significant step for the company in proving the practical value of humanoid robots in real-world logistics. Instead of polished demo clips, this result proves the robot can handle real warehouse tasks every day.

Polymathic: Simulation is one of the cornerstone tools of modern science and engineering

Using simulation-based techniques, scientists can ask how their ideas, actions, and designs will interact with the physical world. Yet this power is not without costs. Cutting edge simulations can often take months of supercomputer time. Surrogate models and machine learning are promising alternatives for accelerating these workflows, but the data hunger of machine learning has limited their impact to data-rich domains. Over the last few years, researchers have sought to side-step this data dependence through the use of foundation models— large models pretrained on large amounts of data which can accelerate the learning process by transferring knowledge from similar inputs, but this is not without its own challenges.

/* */