Robots walking down the street surrounded by astonished onlookers are an increasingly common sight. However, these machines are not yet the all-in-one assistants you would want in a kitchen or factory, and the main bottleneck is data. Like humans, robots learn best through experience. The challenge is that physically teaching these machines to do so many things in different settings is laborious and time-consuming.

“The natural idea is to use simulation as a training ground. While there has been significant progress in the last few years in the physics engines that power robotics simulators, one of the remaining challenges is creating sufficiently rich and varied simulation content to capture the complexities of the real world,” says Russ Tedrake, professor of electrical engineering and computer science (EECS), aeronautics and astronautics, and mechanical engineering at MIT and principal investigator at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL).

It turns out that AI agents, semi-autonomous programs that “think” and perform well-defined tasks, can lend a hand create the realistic virtual settings that robots need. Fresh “SceneSmith” Developed by researchers at MIT CSAIL and the Toyota Research Institute, the system uses three agents to assemble objects, walls and the overall appearance of a 3D scene. Recreated indoor spaces such as restaurants, bedrooms and hotels are more realistic and detailed than previous systems, helping robots practice skills and try out different ways of performing tasks before turning them on. In turn, engineers save time on real-world testing.

Agents have a sense of what everyday places should look like because each of them references a multimodal system called the vision-linguistic model (VLM), specifically the state-of-the-art VLM GPT-5.2. It is trained from a gigantic amount of text and images from the Internet to support more visual cues. This advanced model provides each agent with some kind of spatial knowledge: first the “designer” agent generates the elements of the scene, then the “critic” advises on whether it looks realistic, and finally the “orchestrator” manages their work back and forth, deciding when the design is completed. Once the three VLMs have completed their artistic collaboration, the scene will be ready to be loaded directly into physics simulation software.

“We found that the system could construct 3D scenes in the same way a human designer would,” says MIT EECS graduate student Nicholas Pfaff, a CSAIL researcher and lead author of the study paper with Tedrake presenting the work. “We shot over 1,300 scenes using a leading VLM that has prior web-scale solutions, and created incredibly creative and diverse arrangements. I didn’t teach this system in prompts; it just improvised.”

Talk to my agent

With VLM agents, you can ask SceneSmith to do things like “generate a garage with a car, a workbench, tires stacked in the corner, and a ladder against the wall” and get a virtual playground opulent with items for the robot to tinker with. These rooms are decorated with up to six times the number of items per scene compared to previous methods, making them perfect for helping robots learn skills such as putting a cup in the sink, arranging fruit on plates, and carrying a soda can from shelf to table.

With so many opulent virtual environments at your fingertips, you can assess whether your robot is ready for deployment without having to do a lot of trial and error in the physical world. Researchers tested different roadmaps (also called “rules”) in SceneSmith’s digital worlds, creating 100 unique spaces. A VLM agent evaluated each attempt and found that the robot’s plans were flawed and the machine often failed to perform its duties. Humans agreed with the model’s verdict more than 99 percent of the time, which can lend a hand roboticists eliminate erroneous approaches in simulations before the robot starts moving in the real world.

But how realistic are these virtual worlds? This can be complex to prove completely, so researchers have approached the question from several angles. The most telling test: the politics of a pre-trained robot were thrown into the generated environments – an AI controller trained primarily on real-world data that had never seen a SceneSmith scene. In one test, users told the system to “take an apple out of the bowl and place it on the cutting board,” and the simulated robot did exactly that. If the scenes didn’t closely resemble the actual settings from which the policy was learned, it simply wouldn’t work.

The team remotely controlled robots in virtual spaces, guiding them to open cabinets, put away bottles and move between rooms. Their experiments showed that environments withstand long-term physical interactions beyond visual inspection.

Behind the scenes

Each of the agents SceneSmith uses has a well-defined role in the generation process, completing scenes step by step. They basically create a floor plan and bring it to life.

Let’s say you want to create a scene similar to the first floor of a house. The VLM “designer” started with a general layout, which the “critic” reviewed and then the “orchestrator” signed. Agents repeat this approach at every stage: adding furniture, placing objects on walls, then ceilings, and finally dropping objects that can be manipulated by robots. For example, VLMs can add cabinets that robots can open and close – an articulated element that was often absent in previous baseline solutions.

At each stage, the second VLM ensures that the scene is practical, advising, for example, the removal of the bathtub from the living room. The third VLM ensures that a high-quality scene is generated, even reversing the design process by a few revolutions if the graphics are not up to par. Once the three VLMs have completed their artistic collaboration, the mechanics of the physical world will be added using simulation software.

With a thorough understanding of what rooms should look like, where objects should be placed, and real-world physics, SceneSmith has a noticeable advantage over previous methods. Compared to scene generation baselines such as “HSM“And”Holodeck”, SceneSmith created environments with more facilities, including a private office, a pottery shop, and even a Minecraft-inspired game room.

SceneSmith was also a favorite of over 200 users. They found that the system’s graphics were more realistic more than 90 percent of the time. They also observed that, overall, this approach followed directions more accurately than other approaches. In other words, it was best at generating virtual playgrounds that users actually wanted to see.

Multi-talent system

Realism, variety and richness are SceneSmith’s strengths, even when it comes to generating individual 3D objects. You can ask it to create a wheeled cart and it will create a 2D image, which it will then turn into a detailed model with physical properties such as mass, friction and inertia.

However, such a detailed process requires a trade-off in speed. Creating a single scene can take hours as agents create and carefully analyze each object. With more processing power, the system could see a dramatic increase in performance. CSAIL engineers also hope to expand into deformable objects (such as sponges) if extensive 3D libraries become available.

“SceneSmith represents a significant advance in this area by providing an agent-based framework for generating simulation-ready indoor environments with a simple text message,” says Jeremy Binagia, an applied scientist at Amazon Robotics who was not involved in the research. “It advances the state of the art in several ways, including pushing the boundaries of object density in a simulated environment, ensuring that all objects are physically accurate (not just visually realistic), and creating assets that are not limited to a fixed library because they can be generated using text-to-3D.”

Pfaff and Tedrake wrote the paper with Thomas Cohn SM ’24, an MIT graduate student and CSAIL researcher; and Toyota Research Institute roboticists Sergey Zakharov and Rick Cory SM ’08, PhD ’10. Their work was supported in part by Amazon, the U.S. Office of Naval Research, the Toyota Research Institute, and the U.S. National Science Foundation.

The team brought their findings into the spotlight at last week’s International Machine Learning Conference.

The post AI agents create virtual playgrounds to lend a hand robots obtain key training data appeared first on AI SCKOOL.

Loons, gulls, puffins and petrels are just some of the 100 species of birds that can both fly and swim. These diving birds can dive into the water to swim to their prey, then rise into the air and fly away.

Inspired by these naturally aquatic fliers, engineers at MIT and EPFL in Lausanne, Switzerland, designed a robot that can swim underwater, then emerge from the water and continue flying in the air, much like diving birds.

The “Flapping Wing Air-Aquatic Vehicle” (FAAV) weighs just under 300 grams (about half a pound) and is intended to lend a hand scientists study the mechanics that enable diving birds to fly through air and water.

The robot has a central body, or fuselage; two malleable, flapping wings; and steerable tail. The wings and tail can be exchanged for different sizes. In experiments conducted in a reservoir and at a local lake, engineers identified combinations of wing size, flapping frequency and tail angle that enable the robot to seamlessly transition from swimming through water to piercing the surface to flying through the air.

The post The recent flapping robot swims and flies like a diving bird appeared first on AI SCKOOL.

In recent AI and Society Forum at MITexperts from across the Institute discussed the potential benefits and risks of technological innovations on work, the nature of work, civic discourse, election administration, and other topics.

The event included individual research presentations and panel discussions as well musical performance explores the application of generative artificial intelligence in art.

The forum was co-organized by School of Humanities, Arts and Social Sciences (SHASS) i Social and ethical responsibilities of computing (SERC). It was presented in collaboration with two strategic MIT initiatives: MIT Generative Impact AI Consortium (MGAIC) i Collaborating on Human Insight at MIT (MYTHICAL).

Agustín RayoKenan Sahin, Dean of SHASS i Then HuttenlocherDean of the MIT Schwarzman College of Computing, delivered opening remarks.

Rayo said bringing together scientists from across MIT was intentional because understanding the impact of artificial intelligence requires expertise from a variety of disciplines across the institute.

“Paying attention to the social implications of AI does not mean deviating from MIT’s mission; it is a way to ensure that our technical leadership has maximum impact,” Rayo said.

Huttenlocher added that the rapid development of computer science and artificial intelligence makes supporting interdisciplinary conversations and research crucial.

“Understanding where AI excels and where it falls short is essential not only to unlock its benefits, but also to avoid critical errors, over-reliance and unintended consequences,” Huttenlocher said.

Work and artificial intelligence

The forum, held on May 12 in the Tull Concert Hall in MIT’s Linde Music Building, began with a keynote speech by an economist Author DavidDaniel (1972) and Gail Rubinfeld Professors in the Department of Economics at MIT. The author challenges the common narrative that AI will simply eliminate jobs, proposing instead that the impact of technology depends on how it affects the scarcity and value of human expertise.

“When I think about how technology affects the value of work, I think about it in terms of how it changes the scarcity of expertise, whether it makes it more valuable or whether it makes it more of a commodity,” he said.

The author stated that what matters is whether automation removes routine support tasks or expert tasks. He argued that artificial intelligence is likely to create novel specialized jobs, requiring proactive policies around worker training, wage insurance and broader capital ownership.

This was followed by a panel discussion moderated by Rob Loughlin, partner at McKinsey & Company, featuring experts from MIT who discussed how work is changing and what this means for society.

Daniel RusPanasonic professor of computer science at MIT and director of the Computer Science and Artificial Intelligence Laboratory (CSAIL), described his excitement about the ways artificial intelligence can improve the workplace.

“I would like to imagine the robot as your friend and assistant, someone who watches you and wonders how to help you, as a person to whom you can set high-level tasks,” she said.

Still, Rus said, human judgment remains key to decision-making.

“We could really think about working with artificial intelligence tools, but the human role as the decision maker, the rational thinker, the person deciding the next step, whatever it may be, remains extremely important,” she said.

David Mindellprofessor Aeronautics and Astronautics and Dibner Professor of the History of Engineering and Manufacturing in the Science, Technology and Society program, say the nature of work has constantly changed over the years, but “it’s the new work that counts.”

“We need to support individuals, the economy, the professions to constantly create new work,” he said. “It is absolutely essential that we give young people the tools and allow them to do what they find creative and show us what their new job will be.”

Panelists also discussed the need to maintain safety standards while exploring ways to find improvements. Mindell gave the example of cargo flights, which require six pilots due to the length of the flight.

“We don’t yet know how to reduce this number from six to five, much less to two, one or zero. There is a lot of money behind solving this problem, but there is also a very rich system that has evolved to ensure the security of these systems,” he said.

Sendhil MullainathanProfessor Peter de Florez, a dual faculty member in MIT’s economics, electrical engineering, and computer science (EECS) departments, described a vision of the utility and development of artificial intelligence that delivers productivity improvements, but also cautioned: “I think it’s very useful to distinguish productivity gains from the factors that actually drive long-term growth.”

Either way, Mullainathan said, it’s clear we’re entering a period of wide divergence in AI’s impact on the workforce.

“If you ask, ‘How exactly will organizations restructure?’ I don’t know. But will there be a lot of restructuring? It’s hard to believe that there won’t be many restructurings. And in some sense, if we know that we’re entering a period of high volatility, that in itself is extremely instructive,” he said.

Democracy and artificial intelligence

The second session of the day focused on artificial intelligence technology and its impact on democracy.

Chara Podimata’42 assistant professor of career development and assistant professor of operations research and statistics at the MIT Sloan School of Management, presented her research on auditing vast language models for bias in election information.

“Algorithms now decide many things about our lives,” she said. “When it comes to chatbots and election information, if I take two people and they are interacting with the same chatbot… how will the chatbot respond? How will it personalize the information it gives to these people?”

A longitudinal study of 12 major models conducted during the 2024 U.S. presidential election season found that responses varied significantly depending on the demographics and political leanings provided. Her research team is currently working on a novel audit of the 2026 U.S. midterm elections, using a redesigned survey with input from political science experts.

During a panel discussion moderated by Songyee Yoon, founder and managing partner at Associate Venture Partners and member of the MIT Corporation, experts expressed concerns about AI’s potential to erode democratic norms and processes, but also explored the potential positive impacts.

Bailey FlaniganTheodore T. Miller (1922), a career development professor in the Department of Political Science who holds a joint appointment at the MIT Schwarzman College of Computing with EECS, stated that she was skeptical of the way some people exploit artificial intelligence as a tool that can get people to make decisions or reach consensus more quickly.

“And there’s reason to think it’s nice because it’s more effective. It’s easier. But it loses a lot of those procedural elements of democracy, which are the rituals in which we meet and make decisions,” she said. “And I think it’s a mistake to forget about that when we start thinking about automation.”

Charles Stewart IIIKenan Sahin (1963) distinguished professor of political science and founding director MIT’s Election Data Lab and Science Labhe said one challenge is that government structures are not evolving at the same pace as technology.

Stewart said his biggest concern is the potential for artificial intelligence to cause chaos during and after the election.

“If and when something goes wrong, it can go really bad, really bad. If the election is contested, it could lead to violence,” Stewart said.

“We have already seen that in the era of low technology, election results are being manipulated. What worries me is what I will see on the upcoming election day and the Wednesday after, and whether artificial intelligence has helped create irreversible disruptions in the electoral system,” he added.

Lily TsaiFord Professor of Political Science and founding director MIT Management Lab (MIT GOV/LAB), in many ways artificial intelligence is contrary to the democratic norms and commitments necessary for a fit democracy.

“It’s really important not only in terms of design principles, but also the obligations of designers to know the values ​​and principles that characterize democracy: agency, political equality, mutual respect, inclusion and autonomy,” Tsai said.

Tsai also noted that her research has shown that some people feel more comfortable interacting with machines. She described a “Socratic dialogue chatbot” her team designed that asks people to express the thinking behind their beliefs and positions.

“And that, interestingly, seems to be softening their political stance in this process,” Tsai said. “So there are absolutely examples of ways in which AI can have a positive impact on democracy. But it’s really about designing based on the right principles and rigorously evaluating them.”

The post Exploring the social impacts of artificial intelligence appeared first on AI SCKOOL.

The Hertz Foundation announced that it has awarded 2026 scholarships to three current MIT students and an incoming Ph.D. These are: Annika Marschner, Alvin Q. Meng, Zachary S. Siegel, and Matthew Wanta.

This prestigious science and technology award provides each recipient with five years of financial support – a stipend and the equivalent of full tuition – giving them extraordinary freedom to conduct groundbreaking research as part of their master’s thesis.

“What particularly impresses me about this cohort is their fearlessness in taking on new challenges and pushing the boundaries of science,” says Philip Welkhoff, a Hertz fellow and director of the Gates Foundation’s malaria program, who co-led the selection process. “Each has demonstrated tremendous creativity, commitment and vision, and I can’t wait to see what each of them will accomplish with the freedom to innovate that the Hertz Fellowship provides.”

In addition to funding, fellows receive lifetime access to Hertz Foundation programs, including events, mentoring and networking opportunities, and since the fellowship’s creation in 1963, more than 1,300 fellows have been named. The connections made between these individuals have spawned joint start-ups, research and commercialization in a range of technology, science and engineering fields. Hertz Fellows have contributed to groundbreaking discoveries in fields such as advanced medical therapies, global defense networks and the James Webb Space Telescope.

This year’s MIT-affiliated recipients are among a total of 19 Hertz Foundation Fellows selected from across the United States.

Annika Marschner ’26 majored in mechanical engineering and will begin her PhD at MIT in the fall. Her undergraduate research focused on the development of novel technologies for both biointerfaces and bioinspired systems, including a custom stereoscope-compatible desktop incubator and extrusion-based desktop bioprinter for the MIT Raman Laboratory, a fiber optic bioprinting system for the ETH Zurich Tissue Engineering and Biofabrication Laboratory, and large-scale hardware designs for robotic systems in the Robotics Laboratory Biomimetic MIT. Marschner’s bachelor’s thesis focused on improving the speed and dexterity of animated limb movements in biologically inspired robots. As a student, she plans to continue working on the design of equipment and control systems in biologically relevant environments, particularly in the areas of assistive medical technologies and surgical robotics.

Alvin Q. Meng is a PhD candidate in inorganic chemistry, focusing on understanding the fundamental interactions underlying chemical structure and reactivity. Currently, he is researching iron-sulfur clusters under the supervision of prof Daniel L.M. Suess. Born in Tianjin, China, Meng immigrated to the United States at the age of 10. He received bachelor’s degrees in chemistry and mathematics from the University of Virginia, where he worked in the research group of Professor W. Dean Harman. His research included the synthesis and characterization of dihapto-coordinated tungsten cyclopentadiene complexes, focusing on a class of unusual binuclear species containing a carbon-carbon bond connecting two five-membered metal-bonded rings.

Zachary S. Siegel is a graduate of electrical engineering and computer science, completing doctoral studies in the field of: Laboratory of Computer Science and Artificial Intelligence, where he works at the intersection of robotics, cognitive science and artificial intelligence. He graduated summa cum laude from Princeton University with a BSE in Computer Science and a minor in Philosophy, receiving honors including Tau Beta Pi, Sigma Xi, and the Award for Outstanding Independent Work in Computer Science. His major thesis, supervised by Tom Griffiths and Jacob Andreas, examined how people infer the goals of others in open-ended, real-world environments. Siegel demonstrated how Bayesian inference serves as an true model for predicting people’s goals by comparing partial observations with a learned library of possible plans weighted by their prior probabilities. The goal of his doctoral research is to build machines that learn and reason more like humans – systems that can learn from circumscribed data and generalize to novel situations by combining robot planning and Bayesian inference. Siegel is particularly interested in combinatorial generalization: the human ability to compose known skills in novel ways to solve previously unseen problems without additional demonstrations. At MIT he is advised by, among others: Leslie P. Kaelbling, Tomás Lozano-Pérez, and Joshua B. Tenenbaum.

Mateusz Wanta is an incoming graduate student who will begin operations research at MIT in the fall. He is a 2026 graduate of the United States Military Academy at West Point with a bachelor’s degree in computer science and mathematical sciences, both with honors. His work focused on machine learning for autonomous systems, integrating probabilistic modeling and computer vision in collaborative drone search and swarm control. In cooperation with the DEVCOM Armament Center, Wanta has developed computer vision models for detecting energy defects in artillery ammunition, enabling quick and non-invasive quality control in defense production. His work with the U.S. Special Operations Command and Army C5ISR organizations focused on autonomous aerial search and detection, where he built simulation architectures for probabilistic target location and multi-agent coordination. Wanta served as a company commander in Bravo Company, 2nd Regiment; president of Upsilon Pi Epsilon; and vice president of Phi Kappa Phi. He is a graduate of the Astronaut and Sapper School, as well as an army officer in the Cybercorps.

The post MIT chapters win 2026 Hertz Foundation Fellowships appeared first on AI SCKOOL.

For remotely operated underwater vehicles, gloomy and rugged waters are often unthinkable. When vehicles sink to the seabed or dig through a sand pit, they can kick up clouds of sediment that obstruct the view of on-board cameras. Often the only thing that can be done is to wait for the sea dust to settle before the vehicle can safely continue its journey.

But a modern underwater mapping technique developed by engineers at MIT and the Woods Hole Oceanographic Institute (WHOI) could allow vehicles to see through murky, low-visibility waters.

This method combines visual images from optical cameras with acoustic data from sonar sensors. This combination allows the vehicle to quickly map the general shape of its surroundings using sonar, even in waters with indigent visibility. The vehicle can move toward specific shapes in a sonar-mapped environment, getting close enough for optical cameras to recognize specific objects in detail.

The technique is similar to combining a dolphin’s echolocation with a sea turtle’s close-range vision, allowing it to see and navigate through murky water in real time.

The researchers tested this method in tank experiments where they could control how apparent the water was. Even in the cloudiest conditions, the system was able to look through the sediment to map the tank surroundings and visualize centimeter-scale details of objects in the tank.

The team continues to refine the technique, which they call Sonar-MASt3R. They predict that the mapping method could safely guide underwater vehicles through turbid environments for a range of applications, including scientific research, construction and maintenance of underwater facilities, and deep-sea mining.

“We hope this work will enable us to perform more operations in difficult, low-visibility conditions and help provide greater coverage in areas where it is currently difficult to operate,” says Amy Phung, a graduate student in MIT’s School of Aeronautics and Astronautics who led the work.

Phung introduced article detailing Sonar-MASt3R this week at the IEEE International Conference on Robotics and Automation (ICRA). The paper was co-authored by Richard Camilli, senior scientist in applied ocean physics and engineering at WHOI.

The best of both

To see underwater, scientists generally apply an either-or approach, using optical cameras or sonar sensors to guide them. Optical cameras can provide detailed images of the scene, but only in waters that are relatively clear and well-lit. In contrast, sonar sensors work equally well in clear and turbid water; by emitting acoustic waves and measuring the time and angle of their return, sonar sensors can determine the exact shape, distance and depth of objects in the environment, although the sonar map does not contain any visual details.

To make the most of both modes, scientists decided to combine them in a modern approach known as “optical-acoustic fusion.” In several previous works, research groups have combined sonar and optical data in mapping techniques that are mainly aimed at object recognition and work environment reconstruction. Most techniques require time to synchronize and process data, so they do not work in real time, and only a few can map the environment in 3D. None of them have been applied to high-resolution underwater mapping in turbid and turbid conditions.

Phung, a student in the MIT-WHOI joint program, and Camilli, her advisor, intended to develop an optical-acoustic fusion technique that could generate detailed 3D maps of underwater environments in real time and in low visibility conditions. The team was motivated in part by the challenges of safely recovering unexploded underwater mines.

“Areas where it is unsafe for ships to be present may contain old explosives, and robotics is the best way to dispose of them safely,” says Camilli. “But many of these explosives are deployed in surf zones, where visibility increases the challenge of doing so safely. This is one of many applications to which our technique can be applied.”

Shadowy, mapable

The modern Sonar-MASt3R method is based on the existing MASt3R technique developed by scientists in France. MASt3R is an image matching algorithm that is trained to take visual images of the same scene and quickly estimate the relative depth of each pixel in the scene. In this way, MASt3R can generate a 3D map of the environment in real time based on 2D images from the camera.

“The disadvantage is the lack of sense of scale,” Phung says. “It will say, ‘this pixel is five units closer than this pixel,’ but it won’t be able to tell if it’s 5 meters or 5 feet.”

Fortunately, sonar provides absolute scale measurements. The time of sonar reflections can be directly translated into the specific depth and distance of the objects from which the signals were reflected, as well as their shape and contour.

In their modern work, Phung and Camilli used sonar data to correct the scaling of MASt3R and generate precise 3D maps of underwater environments. Even in murky water, a sonar-corrected map would enable the vehicle to know the exact location of objects, and therefore the distance to which it should safely approach, for more precise inspection, which the vehicle could then perform using conventional optical cameras.

The team tested Sonar-MASt3R in experiments with a tank filled with water, sediment and various objects such as a miniature boulder, a coffee mug and a packing crate. They also placed a robotic arm inside the tank on which they mounted an underwater camera and a sonar sensor.

For each experiment, a sweep trajectory was first performed in which the robotic arm slowly moved from one side of the tank to the other to capture sonar and visual data. During the first sweep, Sonar-MASt3R quickly creates a rugged sonar-based map of the shapes and contours of the tank and its features. The rugged map is then used to capture close-up camera images of the features, which is used to improve the resolution of the map. The “keyframe” method allows you to quickly compare each modern image frame with the last keyframe. If a frame contains modern information that is not contained in the last keyframe, the image is added to the map as a modern keyframe. If it is similar, it is immediately rejected. In this way, the approach can quickly populate the map with relevant visual details in real time.

The researchers tested their modern approach underwater, testing eight different levels of turbidity, which they created by mixing sediment in a tank. Compared to other optical-acoustic fusion approaches, Sonar-MASt3R generated more exact 3D maps and resolved finer details at the centimeter scale and in cloudier conditions. In the cloudiest conditions, which the robot arm’s cameras were unable to see, its sonar sensors were able to generate a rugged map of hidden objects in the tank. This initial map allowed the arm to navigate safely through the shadowy and approach specific objects, which its underwater camera could then visualize in greater detail.

“An analogy would be walking into a china shop in the dark and trying to find a specific coffee mug without knocking anything over,” Camilli suggests. “That would allow you to do it.”

The team plans to test this approach in natural underwater conditions, where they suspect the mapping task should be simpler.

“It’s like an echo chamber in the tank,” Camilli says. “It’s like trying to do it in a mirror at an amusement park, where all the distortions and echoes and ghost images come in, which really complicates the processing. If you put it in the real world, it should be easier.”

Then, they say, Sonar-MASt3R can aid scientists explore safely in gloomy, murky and murky underwater areas.

“The real value of this effort is that we can use this technology in mission scenarios that are currently not feasible,” Phung says. “There are many missions that can’t be accomplished because we don’t have the ability to observe or perceive.”

This research was supported in part by NASA and the National Science Foundation.

The post Novel imaging system sees through murky water appeared first on AI SCKOOL.

When a team of MIT students showed up at a national robotics tournament, their robot – aptly named Timbot – didn’t work. They were invited to demonstrate Timbot at the launch event United States Governors Cup in Washington, D.C., a March Madness-style competition that featured high school robotics teams from all 50 states.

Solving problems on the fly is a given at robotics tournaments. Timbot had a few technical issues, mainly with Wi-Fi, so the team sat cross-legged on the floor and got to work. Meanwhile, high school students began to gather and ask questions about the cabling and subsystems. After about an hour, Timbot was working againcollecting and throwing foam balls as intended.

“It really was a great moment,” says freshman Lily Sand. “We ended up connecting the robot with a long Ethernet cable instead of using a wireless connection, and many of the students thought, ‘Oh, we do that too!’ It was a nice touchpoint.”

Using a cultural touchstone for good

Getting younger students interested in robotics is one of the goals of MIT students as members of a up-to-date club, FIRSTxMYTHwhich started at the beginning of the academic year. All members are graduates of programs offered by FIRST Robotics (FIRST), a nonprofit organization dedicated to sparking interest in STEM among primary and secondary school students around the world through team robotics programs and competitions.

FIRST has deep roots at MIT. Inventor Dean Kamen worked with the overdue MIT professor Woodie Flowers, a pioneer of hands-on engineering design education, to establish the FIRST Robotics Competition in 1992. The competition was modeled after the inventive robotics competition that Flowers developed for his Iconic Mechanical Engineering Class 2.70 (Introduction to Design)i.e. currently 2,007 (Design and production I).

With FIRST, students learn more than just designing, building and programming robots. The program emphasizes an ethos of “gracious professionalism” – a term coined by Flowers to describe high quality work, respect and cooperation, even in a competitive context. Students also build self-confidence, gain leadership experience, and improve communication skills as well as technical knowledge.

Many FIRST graduates feel a deep sense of gratitude for the program and a mighty desire to remain involved. Debbie Ang, co-founder of FIRSTxMIT, continues to mentor the team at her high school in Modern Hampshire. However, there are few FIRST alumni clubs on campus. Ang and co-founder Perry Han, also a sophomore, met in high school for FIRST and reconnected at MIT. “We noticed that FIRST was founded here, and yet there was nothing organized on campus, even though we kept meeting people who had founded FIRST and still cared about the community,” he explains.

In fact, participation in FIRST is something of a cultural touchstone among MIT students. MIT Associate Director of Admissions Trinidad Carney, a liaison to FIRST Robotics, estimates that 15 to 20 percent of students have participated in the program.

Han and Ang worked with Carney to launch FIRSTxMIT under the auspices Edgerton Centerto strengthen connections among the MIT FIRST community and provide members with the opportunity to channel their passion for FIRST into outreach and public service. Their hunch about the untapped potential of the alumni club was spot on: the kick-off event attracted 185 students, and they have about 200 on their Discord channel.

Sharing the “Power of the FIRST”

Now the club is not operating. They organized a meeting for Modern England FIRST alumni; worked with Josiah Quincy Elementary School in Boston to launch the LEGO Robotics League; volunteered as a judge at local competitions; and helped the MIT Admissions Office get there. Carney, who serves as the club’s advisor, says, “We’ve actually had other universities reach out to us and say, ‘How did MIT manage to launch a club that’s so successful and so exciting?'”

One of the club’s most ambitious undertakings to date was building Timbotin three days, in January, during the Independent Activity Period. Robot in 3 Days (Ri3D) is a collegial challenge in which students build THE FIRST Robotic Competition-level robot in 72 hours, which would take a high school team about six weeks. Experimental roboticsThe consortium, which uses an experiential robotics platform to promote engineering and public service, provided support for MIT’s Ri3D challenge and invited the team to serve as STEM ambassadors at the Governors Cup.

In addition to the robotics competition, the two-day event brought together governors and leaders from government, education, industry and others to highlight the critical role states play in supporting STEM education.

To this end, the FIRSTxMIT team demonstrated Timbot, spoke to high school students, staffed the MIT recruiting booth, and met with VIPs, sharing the value of project-based STEM enrichment opportunities like FIRST. “Having MIT students tell the story of the power of FIRST is incredibly compelling,” says Carney. “They can say, ‘I did this in high school, it made me who I am, and now I’m at MIT still building and giving back.’

Many governors stopped by the MIT recruiting booth to talk to students, including Massachusetts Governor Maura Healey. “She talked about the importance of STEM education in primary and secondary schools and was very supportive,” says Sand, logistics coordinator at FIRSTxMIT.

In addition to inspiring others, MIT students were inspired by the Governors Cup themselves. Han recalls a conversation with a state senator from Ohio, a former teacher and mighty supporter of programs like FIRST. “It really showed me that if you have people in government who are excited about STEM education, it can really be a success.”

Building a better future

Looking to the future, Han and Ang plan to spend some time further refining the club’s organization and future goals. Practical information activities occupy an crucial place in their plans. “FIRST places a strong emphasis on starting new teams, supporting underprivileged communities and spreading awareness,” says Ang. “Many of us feel FIRST has played an important role in shaping our academic and career paths, so we want to give that opportunity to others.”

“Part of our goal is to put the robot in the hands of as many students as possible to kind of give them a sense that STEM is not just about reading an AP Physics C-Mechanics textbook,” Han adds. “It’s actually putting those ideas into action and building something useful.”

They also have no shortage of up-to-date ideas to explore. Han is particularly interested in encouraging students to earn credits through the Undergraduate Research Opportunities Program or coursework credit for projects such as Ri3D, as well as encouraging students in the Gordon Engineering Leadership Program to gain leadership recognition through mentoring the robotics team. He also wants to explore how to leverage FIRST alumni networks to aid students develop professionally.

Regardless of the path they choose, Carney has no doubt that they are making an impact. She saw their full potential when they built Timbot.

“These students, many of whom had never met before, came from different backgrounds: different schools, different regions, with different life experiences,” he says. “But they worked together with respect, curiosity, and generosity. They are collaborative, mission-driven, and passionate about creating opportunities for others. They make MIT better and will make the future better.”

The post STEM ambassadors appeared first on AI SCKOOL.

In the aftermath of a devastating earthquake, unmanned aerial vehicles (UAVs) can fly through a collapsed building to map the scene and give rescuers the information they need to quickly reach survivors.

However, this remains an extremely challenging problem for an autonomous robot, which would need to quickly adjust its trajectory to avoid sudden obstacles while staying on course.

Researchers from MIT and the University of Pennsylvania have developed a modern trajectory planning system that addresses both challenges simultaneously. Their technique enables the UAV to react to obstacles within milliseconds while maintaining a velvety flight path, which minimizes travel time.

Their system uses a modern mathematical formulation that ensures the robot will safely reach its destination along a feasible path, while requiring less computation than other techniques. In this way, it generates smoother trajectories faster than state-of-the-art methods.

The trajectory planner is also powerful enough to enable real-time flight using only the robot’s onboard computer and sensors.

The open-source system, called MIGHTY, does not require proprietary software packages that can cost hundreds of thousands of dollars. It can be more easily implemented in a wider range of real-world settings.

In addition to search and rescue, MIGHTY can be used for applications such as last-mile deliveries in urban spaces, where UAVs must avoid buildings, wires and people, or for industrial inspections of convoluted structures such as wind turbines.

“MIGHTY achieves comparable or better performance by using only open source tools, which means that any researcher, student, or company – anywhere in the world – can use it freely. By removing this cost barrier, MIGHTY helps democratize high-performance trajectory planning and opens the door to a much broader community that can leverage this work,” says Kota Kondo, an aeronautics and astronautics graduate student and lead author of the paper on this planning tool trajectory.

In the article, Kondo is joined by Yuwei Wu, a graduate of the University of Pennsylvania; Vijay Kumar, professor at UPenn; and senior author Jonathan P. How, Ford Professor of Aeronautics and Astronautics and principal investigator at the Laboratory for Information and Decision Systems (LIDS) and the Aerospace Control Laboratory (ACL) at MIT. Tests appears in .

Overcoming compromises

When Kondo was a child, the Fukushima Daiichi nuclear accident occurred after the Great East Japan Earthquake. After school was canceled, Kondo stayed home and watched the news every day while workers surveyed and secured the reactor site. Some workers still had to enter hazardous areas to contain the damage and assess the situation, exposing them to high doses of radioactive materials.

“I was passionate about creating autonomous robots that can enter dynamic and dangerous situations and then come back and report back to humans who stay out of danger,” Kondo says.

This task requires a good trajectory planning tool, which is software that determines the path the robot should follow to get from point A to point B safely.

However, many existing systems impose trade-offs that limit performance.

Although some commercial systems can quickly generate velvety trajectories, they can cost hundreds of thousands of dollars. Open source alternatives are often weaker compared to commercial solutions or challenging to exploit.

With MIGHTY, Kondo and his colleagues have developed an open-source system that generates high-quality, velvety trajectories by responding to obstacles in real time and that runs speedy enough to fly using only on-board components.

To do this, they overcame a key challenge that limits many open source systems.

These methods usually first estimate how long it will take the robot to get from point A to point B. Based on the estimated travel time, the planner finds the best way to reach the destination.

Although using a fixed travel time allows the planner to quickly generate a trajectory, it has drawbacks. First, if the UAV needs to move very far away from obstacles to avoid obstacles, it may have to boost its speed to meet the set travel time budget. This makes it challenging to avoid sudden threats.

A STRONG method

Instead, MIGHTY uses a mathematical technique called the Hermite spline, which optimizes travel time and flight path in one step, creating a velvety trajectory that can be precisely controlled.

“Combined optimization of the spatial and temporal components allows us to get better results, but now the optimization is so large that it is more difficult to solve in the possible time,” Kondo says.

The researchers used a clever technique to reduce this computational overhead.

Instead of generating a trajectory from scratch every time, MIGHTY pre-guesses the trajectory. It then refines the trajectory through iterative optimization using the scene map generated by the UAV’s lidar sensors.

“We can reasonably guess what the trajectory should be, which is much faster than generating the whole thing from nothing,” Kondo says.

This allows MIGHTY to react in real time to unknown obstacles, while maintaining a velvety trajectory and minimizing travel time. The system uses on-board UAV components, which is crucial in applications where the robot can move far from the base station.

In simulated experiments, MIGHTY required only about 90 percent of the computational time required by state-of-the-art methods and safely arrived at the target about 15 percent faster than those approaches.

When they tested the system on real robots, it reached speeds of 6.7 meters per second, avoiding every obstacle that came in its path.

“With MIGHTY, everything is integrated into one whole. There is no need to communicate with any other software to find a solution. This allows us to work even faster than some commercial solutions,” says Kondo.

In the future, researchers want to improve MIGHTY so it can be used to control multiple robots at once and conduct more flight experiments in challenging environments. They hope to continue to improve the open source system based on user feedback.

“MIGHTY makes an crucial contribution to agile robot navigation by rethinking the trajectory representation itself. Hermite splines have already been used successfully in visual simultaneous localization and mapping, and it is nice to see that their advantages are now being exploited for trajectory planning in mobile robots. By enabling joint optimization of path geometry, time, velocity and acceleration while retaining local control over the trajectory, MIGHTY gives robots greater freedom to calculate speedy, dynamically feasible movements in cluttered environments,” says Davide Scaramuzza, professor and director of the Robotics and Perception Group at the University of Zurich, who was not involved in this research.

This research was funded in part by the U.S. Army Research Laboratory and the Defense Science and Technology Agency in Singapore.

The post Recent research allows the robot to chart a better course appeared first on AI SCKOOL.

MIT student Sunshine Jiang ’25 and Rupert Li ’24 are the recipients of this year’s Knight-Hennessy Scholarship. Now in its ninth year, this highly competitive scholarship provides financial support for graduate study at Stanford University for up to three years.

Sunlit Jiang ’25

Sunshine Jiang of Hangzhou, China, graduated from MIT in 2025 with a bachelor’s degree with a double major in physics and electrical engineering and computer science, as well as minors in mathematics and economics. He will graduate this month with a master’s degree in engineering and will begin doctoral studies in computer science at the Stanford School of Engineering in the fall.

Jiang is engaged in artificial intelligence and robotics research, developing productive and adaptive data-efficient systems for general-purpose robots that augment accessibility. She has presented her research at major conferences, including the Conference on Robotic Learning, the International Conference on Robotics and Automation, and the International Conference on Learning Representations.

Jiang led the development of AI-powered systems that provide access to customary Chinese arts in rural classrooms, founded cross-country programs that expand girls’ access to STEM education, and created a Covid-19 document amplifying community voices that was featured on China Daily.

Rupert Lee ’24

Rupert Li of Portland, Oregon, is currently pursuing a PhD in mathematics at the Stanford School of Humanities and Sciences. He graduated from MIT in 2024 with a bachelor’s degree, double majoring in mathematics and computer science, economics and data science. In addition to his bachelor’s degree, he also earned a master’s degree in data analytics. Li then traveled to the UK as a Marshall Scholar, where he obtained a master’s degree in mathematics from the University of Cambridge.

Li’s research interests include probability, discrete geometry, and combinatorics. He enjoys serving as a mentor for MIT PRIMES-USA, the high school mathematics research program, and was previously an advisor for Duluth REU, the undergraduate mathematics research program. In addition to the Knight-Hennessy and Marshall Scholarships, he received the Hertz Fellowship, the PD Soros Fellowship and the Goldwater Scholarship, and received the distinction of receiving the Frank and Brennie Morgan Award.

The post Two from MIT have been named 2026 Knight-Hennessy Scholars appeared first on AI SCKOOL.

The post It took technology 40 years to catch up with this zipper design appeared first on AI SCKOOL.

Under a microscope, a bouquet of lollipop-like structures, each smaller than a grain of sand, gently undulates in a Petri dish of liquid. Suddenly they snap together like the jaws of a flycatcher as the scientist waves a miniature magnet over the plate. What was previously a collection of small passive structures immediately transformed into an energetic robotic gripper.

The Lollipop gripper is one demonstration of a modern type of cushioned magnetic hydrogel developed by MIT engineers and their collaborators at École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland and the University of Cincinnati. In the study will appear in the magazine today MIT team reports a modern method for printing and producing a gel that can be transformed into convoluted, magnetically activated three-dimensional structures.

The modern gel could form the basis of cushioned, microscopic, magnetically responsive robots and materials. Such magnobots could be used in medicine, for example to release drugs or take biopsies under the influence of an external magnet.

Setting objects in motion using magnets is nothing modern, at least on a macro scale. For example, we can wave a fridge magnet over a stack of paper clips, which will follow the magnet in response. At the microscale, scientists have designed a variety of magnetic “micro-swimmers” – pieces smaller than a millimeter that can be remotely guided by a magnet to squeeze through miniature spaces. For the most part, these designs work by mixing magnetic particles with a printable resin and attracting the entire float toward an external magnet.

The post With the wave of a magnet, microscopic “magno-bots” perform convoluted maneuvers appeared first on AI SCKOOL.