Quick Look
Iowa State researchers are working to develop and test new methods that could help advance the physical intelligence and capabilities of humanoid robots while also prioritizing the development of safety standards that will allow robots and humans to interact safely.
AMES, Iowa - You may not remember it, but odds are you took a few tumbles during your toddler era. You weren't alone. Falling, after all, is a natural consequence of learning to crawl, walk, climb and jump. Our balance, coordination and motor skills are developing throughout early childhood.
But it also doesn't take long for these abilities - also known as physical intelligence - to become second nature for most, including deceptively complex actions such as walking, grasping objects and navigating our way across a room without having to think about it.
"As humans, we often take our physical intelligence for granted because it becomes so automatic when we're still young," said Bowen Weng, roboticist and assistant professor of computer science at Iowa State University. "But the truth is, it's remarkable. The fact that you haven't fallen over much since you were 3 years old is remarkable. Once humans master a skill, the physical intelligence needed to perform that skill becomes a background process, and our minds are freed up to focus on other tasks."
For humanoid robots, however, physical intelligence is anything but automatic or easy to learn.
"Even with significant advances in AI, the physical design and adaptability of robot bodies are still significant barriers to real-world performance," Weng said. "Robots struggle with physical intelligence because it requires adapting to unpredictable environments, integrating sensory feedback in real time and mastering complex motor skills. These tasks are largely intuitive for humans but notoriously difficult to replicate in machines."
That's why Weng, along with several graduate students in computer science, are working to develop and test new methods that could help advance the physical intelligence and capabilities of humanoid robots while also prioritizing the development of safety standards that will allow robots and humans to interact safely.
The challenge of adaptation
Humanoid robots are designed with a body shape that resembles the human form and are built to interact with human environments and tools. They can be used for a variety of purposes, including assisting humans with tasks, conducting research, or performing dangerous or tedious jobs.
"To help move our society and humanity forward, we have to get more done, do more things and find a way to accomplish it all with greater speed and efficiency," Weng said. "Automation is a step to help us reach that goal, but at the same time, it's important to acknowledge certain fears people might have and explain why humans can't and won't be replaced by humanoid robots."
The challenge, Weng said, isn't replacement but adaptation.
"Humanoid robots need humans because they rely on us for design, training, supervision, ethical guidance and emotional context," he said. "They can't replicate or understand these functions on their own, so our willingness to adapt and utilize AI to strategically support and augment our own abilities is really important."
Weng also noted that new career paths and jobs will emerge in the areas of AI oversight, ethics, design and maintenance as we adapt to human-AI collaboration.
"These will be new opportunities to grow our workforce and economy," he said.
Collaboration, not competition
At the end of a winding back hallway on the first floor of Atanasoff Hall on the Iowa State campus, there's an unassuming door secured with a sturdy keypad lock and a bit of mystery. On the other side is a space filled with cutting-edge possibilities: the ISU computer science robotics lab.
Currently, the lab is home to two highly advanced legged humanoid robots, with the largest standing about 6 feet tall and the smaller reaching roughly the height of a 10-year-old child, as well as a quadruped robot with dog-like attributes and the build of a beagle.
It's also where Weng and his team of student researchers can often be found working with computer software and hand-held controllers, much like those used with video-gaming systems, to help advance the physical capabilities of these robots. This includes taking the humanoid robots through commands to stand up from a lying position, sit down, walk, turn around, wave their arms and shake hands, and directing the dog-like quadruped robot to stand, sit, leap, shake hands, and walk forward, backward and sideways on command.
"Things that humans are able to do easily, a humanoid robot is not able to do that," said Zaid Mahboob, a doctoral student in computer science. "My motivation is help make these robots capable of doing more of these things with greater efficiency."
Mahboob said the group's shared goal isn't to create competition between humans and humanoid robots, but instead to improve the precision, accuracy and speed of the robots so they can safely and effectively collaborate with humans.
As humans, we often take our physical intelligence for granted because it becomes so automatic when we're still young. But the truth is, it's remarkable.
Bowen Weng, roboticist and assistant professor of computer science
Leading with safety
Weng recently co-authored the paper "Repeatable and Reliable Efforts of Accelerated Risk Assessment in Robot Testing" and presented its findings during the 2025 IEEE International Conference on Robotics and Automation. The study aimed to enhance existing accelerated testing frameworks for robots by proposing a new algorithm that's both repeatable and reliable while also assessing a robot's risk of instability from disturbances caused by frontal impacts.
"It's exciting to see robots work," Weng said. "But at the same time, there isn't nearly enough effort being put into responsibly and safely achieving these developments."
Weng's research showed that the new algorithm works reliably and with high probability, and that future testing with other robot types is promising. However, more safety measures are still needed, he said.
"This research makes an 'indirect' impact towards a safer robot," Weng said. "We did not make any contribution to enhance the robot's specific capabilities. We did do something to enhance the test algorithm's specific capabilities, which can be used against the robots someday."
A second study, "Experimental Evaluation of Commercial Quadruped Robots: Stability and Performance in Non-inertial Environments" co-authored by Weng in collaboration with researchers at the University of Massachusetts at Lowell and Purdue University, found that while the quadrupeds showed keen capabilities, challenges with accurate body positioning during aggressive ground motion remain.
This year, the National Institute of Standards and Technology (NIST) also awarded Weng a grant to fund a new research project titled "Testing of Trustworthy Mobility: Standardized Performance Evaluation of Legged Robots Stability."
"Mobile robots that feature limbed structures - particularly legs - are becoming increasingly relevant because of how versatile and adaptable they can be across many different environments and applications," said Weng, noting that legged robotic systems have shown significant promise in applications such as search-and-rescue missions, health care assistance, manufacturing, disaster response and enhanced mobility solutions.
Weng said the new research project will contribute to the improved reliability and effectiveness of legged robotic systems by advancing evaluation standards and by promoting transparency regarding the capabilities - and limitations - of these systems.
"Transparency is vital for fostering public trust, enhancing safety and helping us have meaningful discussions around the practical deployment and responsible use of robots," Weng said.
Keeping it real
Among the graduate students working with Weng, having the chance to conduct research using real robots is frequently cited as a key selling point in their decision to come to Iowa State.
"Most of the robotics labs are just interested in simulation and that kind of thing, but in (Iowa State's) lab, we are conducting real experiments using real robots to get real data," said Yuija Chen, a doctoral student in computer science.
Dylan Khor, a second-year master's student in computer science, said he was also motivated by the opportunity to work closely with Weng.
"I first took (Dr. Weng's) class on introduction to machine learning, and I really liked the way he taught the class," Khor said. "He brought so much energy and knowledge to the subject, and the class was very interactive. So, when I decided I wanted to do research in robotics, I reached out to him right away and here I am. It's been a wonderful experience."
While Weng's focus is on developing standardized safety measures for robots, he also finds great purpose, enjoyment and inspiration in helping guide students in their research. In addition to working with graduate students, Weng leads several undergraduate computer science teams in programming work with legged robots and robotic arms.
Weng said he expects research opportunities will keep growing when the new robotics education lab is fully up and running in the lower level of Durham Hall. This education lab, which is slated for completion during the current academic year, already includes eight robotic arms for research and teaching purposes.
Adoption outlook
Weng said humanoid robots still face some major hurdles to widespread adoption, including high development and unit costs, a lack of standardization, infrastructure gaps and limited use cases, and there are social and ethical challenges that need to be addressed as well.
However, Weng said these challenges and growth barriers also make ongoing research to improve the safety and efficacy of robots even more important.
"The bottom line is, you have to be able to trust it," Weng said, "and the path to proving the trustworthiness of humanoid robots is through human-led research."
That research is happening now at Iowa State.
