When you click on a sponsoring school or program advertised on our site, or fill out a form to request information from a sponsoring school, we may earn a commission. View our advertising disclosure for more details.
Dr. Jing Xiao is a professor and head of the department of robotics engineering at Worcester Polytechnic Institute. She earned both her MS and her PhD in computer, information, and control engineering at the University of Michigan, Ann Arbor.
Dr. Xiao’s research sits at the intersection of computer science and engineering, spanning robotics, haptics, and multi-modal perception. Her work with robotics focuses on two highly related themes: the contact and interaction between a robot or a part/tool it holds and the environment, and the real-time adaptiveness of robots to uncertainty and uncertain changes in an environment based on perception.
“We’ve seen great advances in humanoid robots and their capabilities,” Dr. Xiao says. “Better actuators, battery lives, and control have contributed to better robot hardware. AI-powered perception and motion control let the robot do imitation learning of movements from humans, which, along with reinforcement learning, has contributed to greater capabilities, particularly in locomotion.
To mark the start of the Year of the Fire Horse, the 2026 Spring Festival Gala in Beijing hosted the world’s first fully autonomous martial arts performance (LiveScience). Unitree Robotics’ G1 and H2 machines performed complex acrobatics, wielding weapons. Their expert backflips and perfectly timed tumbles were made possible, in part, by large-scale training in simulation: optimizing control policies in a simulated environment before deploying them on real robots (a process known as sim-to-real transfer). That approach saves money and time.
But the world doesn’t need fully autonomous martial arts performances. What it does need, at least in the dreams of many businesspeople, is a cheap and reliable labor source. In 2024, roughly 76 percent of supply chain and logistics leaders surveyed were experiencing notable workforce shortages; 37 percent of respondents categorized their shortages as high to extreme (Descartes Systems Group). As actuators and controls continue to improve, along with AI perception, humanoid robots could fill the gap. Tesla’s Optimus, Figure AI’s Figure 01, Agility Robotics’ Digit, and Boston Dynamics’ Atlas are betting on it.
“A lot of companies are considering applying humanoid robots in more useful scenarios, like factories and warehouses,” Dr. Xiao says. “These robots are not so much programmed: they are more like embodied AI systems.”
Foundation models, which are large AI systems adapted for robots, are an enormous level-up. They can empower robots to interpret natural language, perceive their environment more effectively, and reason about actions before taking them. Historically, the most successful robots have been highly specialized to a specific task or domain: consider the industrial robot arm, the Roomba, and the da Vinci surgical system.
Foundation models for robots enable greater versatility, adaptability, and trainability. If before a robot was an extremely sophisticated Phillips head screwdriver, it’s now become something much more: a walking, and sometimes talking, physical AI system, what’s known as embodied AI.
But humanoids aren’t perfect yet. They still face engineering challenges in reliability, dexterity, and perception. Tactile sensing has improved, but fine manipulation by robots, in contact-rich environments, remains a major bottleneck. Tasks that many humans perform effortlessly—like handling delicate objects, adjusting grip strength, or sensing subtle changes in texture and force—can befuddle even the most advanced walking, talking AI system.
“Whether humanoids should be used for all kinds of situations, I don’t think so,” Dr. Xiao says. “It’s a question of whether you want to have a really versatile, all-in-one kind of robot that’s humanoid, or whether you want to distribute its capabilities across different devices.”
Human environments were built for human bodies: stairs, things on shelves, items meant to be reached for and grabbed. Fully participating in the workforce often requires whole-body motion, and that’s the aspiration of the humanoid robot. But the gap between what’s technically demonstrated in pilot programs (or in fully autonomous martial arts performances) and what is commercially viable at scale remains wide (McKinsey).
Meanwhile, non-humanoid robots have been enormously successful, with real deployments, measurable outcomes, and increasing usage to show for it. Kiva Systems (acquired by Amazon) has one of the largest robotics deployments in the world, with mobile warehouse robots moving shelves or inventory around fulfillment centers. Ocado’s grocery warehouse is one of the most sophisticated on the planet; its robotic arms can pack over a third of grocery orders in some facilities. Operating in semi-structured spaces, with specific tasks, non-humanoid robots offer reliability, efficiency, and a clear ROI.
“We probably don’t consider a washing machine a robot, but it is,” Dr. Xiao says. “It’s just not high-dimensional. Similarly, we have vacuum robots—we don’t need a humanoid to hold a vacuum cleaner for that. We have better solutions.”
A major challenge for humanoid robots is the expectation that’s being set for them. If we only expect them to do gymnastic movements, then they’ve succeeded. But if we want them across the spectrum of the workforce, say, to help elderly people at home, Dr. Xiao suggests, there are still many gaps to fill. Humans often take for granted the enormous complexity of their bodies: how a single finger has several degrees of freedom and multiple ways in which it can be manipulated. That’s nearly impossible to program.
“We need more data,” Dr. Xiao says. “Visual data and language data are so vast, it’s almost free. But manipulating data is much rarer. There are companies hiring people just to teleoperate robots, to do different kinds of manipulation tasks, and then collect the data, but that’s a very slow and expensive process.”
More broadly, a major engineering challenge is equipping robots—both humanoid and non-humanoid—with the tools they need to function safely and predictably in the real world. Reliability is in many ways a robot’s prime directive, but engineers are still struggling with edge cases, long-horizon tasks, and adapting predictable autonomous objects to the general messiness and variability of everyday life.
“We need to solve problems like reliability in the real world,” Dr. Xiao says.
A key part of this is human-robot interaction (HRI), the field of robotics that studies how humans and robots communicate and coexist. The idea is to create a safe, predictable, and intuitive relationship between man and machine.
Things can go horribly wrong: in 2015, a robot crushed a worker to death at a Volkswagen factory (the company determined it was due to ‘human error’); in 2021, a robot at a Tesla factory pinned an engineer with its claws, causing lacerations (the engineer escaped and fell down a garbage chute, leaving a trail of blood); in 2023, an industrial robot crushed a worker to death at a vegetable packing plant in Goseong, South Korea (the robot’s sensors were designed to identity boxes, and security camera footage indicated the worker had moved near the robot with a box in his hands).
Notably, none of the robots in these violent events were powered by modern AI systems—instead, they were conventional industrial machines operating with preprogrammed control logic and sensors.
HRI is a two-way street. On one side, it involves explainable robot behavior and intent prediction. But on the other hand, it informs how humans can help guide robots to make better decisions. Humans take their perception for granted; robots must learn everything from a blank slate. As Dr. Xiao notes, if a human asks a robot to pick up the cup from the table, a robot needs to know what and where a table is, what and where a cup is, etcetera. But robots are good students: if a human points and says, “Pick up that cup, the silver cup,” it’s speedy and effective at completing the task.
“Human-robot interaction not only is necessary if robots are to serve people in human-centered environments, but can help a robot to achieve a lot of things it cannot achieve autonomously,” Dr. Xiao says.
The future of robotics engineering is arriving quickly: more AI-driven systems, wider warehouse automation, and early humanoid pilots in logistics. To meet the challenges of that future, the next generation of engineering students will need the fundamentals—calculus, probability, optimization, dynamics—but also an understanding of what’s driving the next generation of robotics.
“A roboticist needs to be AI-driven,” Dr. Xiao says. “AI-driven skills, such as knowing neural networks for computer vision perception, transformers for multimodal modeling, and vision-based language models—you need to understand those and know how to use them.”
Soft skills are also important. This is a team-based field that reaches across disciplines, and communication between hardware and AI teams is crucial. And there’s always the question of ethics—a complex issue when considering autonomous devices. Robotics programs like the one at WPI include a focus on ethics in the classroom and emphasize, overall, a strong sense of systems-level thinking.
“Robotics is an extremely interdisciplinary field,” Dr. Xiao says. “A robot has kinematics, actuators, dynamics, control, sensing, software, data management, and decision-making. System-level thinking is exceedingly important. A humanoid robot is, itself, an example of a system: so many different parts and functionalities, all in one entity.”
Matt Zbrog is a writer and researcher from Southern California. Since 2018, he’s written extensively about a wide range of engineering disciplines, with a particular focus on the potential impacts of major technological breakthroughs. His work largely centers around conversations with engineering faculty at top universities, highlighting their research in areas such as nuclear power, artificial intelligence, soft robotics, and semiconductor design. He’s also worked with leaders and subject matter experts at the Institute for Electrical and Electronics Engineers (IEEE), the California Department of Water Resources (DWR), the National Society of Professional Engineers (NSPE), and the Society of Women Engineers (SWE).
By participating in a high school engineering program, students can build a solid foundation in STEM, form professional networks, and gain a clearer sense of their academic and career paths.
Engineers might be the only group of people where you can give them a problem—and they can consider it a gift. The engineering mind thrives on hunting for elegant solutions to complex tasks. That doesn’t mean you should give an engineering student a homework assignment for the holidays, but it does mean you can have some fun with the gift you eventually select.
Engineering internships are an increasingly important part of the transition from student to engineer. Internships provide an opportunity to put theoretical skills to work in hands-on environments. They also give engineering students valuable work experience, networking opportunities, and future career options.
Not long ago, self-driving cars were science fiction. Today, not so much. Influential companies like Tesla, Uber, Apple, and Google boast dynamic auto-drive programs, and many new startups are following their lead.
The 12th annual National Robotics Week (RoboWeek) takes place April 2-10, 2022. Established by Congress in 2010, this tech-focused week is about demonstrating the positive societal impacts of robotic technologies, and inspiring students of all ages to pursue careers related to Science, Technology, Engineering, and Math (STEM).
