When people collaborate in pairs or groups, they can tackle a broader range of challenges and often complete tasks more quickly and efficiently than they could alone. To provide similar support, robots need to quickly understand human actions and instructions, anticipate what their partners will do, and use those predictions to plan and carry out useful actions accurately.
Researchers from GenerativeBionics, the Italian Institute of Technology (IIT), and the University of Manchester have developed a framework aimed at helping engineers build robots that can interact with humans more safely, efficiently, and flexibly. The approach, detailed in a paper published in Nature Machine Intelligence, has already been applied to the development of a new humanoid robot known as ergoCub.
“Our research emerged from years of experience designing successive generations of humanoid robots,” Carlotta Sartore and Daniele Pucci, the study’s first and senior authors, respectively, told Tech Xplore. “Throughout this work, we noticed that robotic hardware was frequently regarded as something fixed, leaving the control system to compensate for its physical limitations.
“This approach could turn the robot’s physical structure into a limitation instead of a foundation for intelligent behavior, slowing down and complicating the development of new humanoid abilities. The problem became especially clear when robots were required to physically collaborate with humans.”
Rethinking humanoid robot design
When developing robots, computer scientists typically account for human partners when determining how a robot should behave or plan its actions. However, they generally design the robot’s physical structure separately, without considering the specific needs, abilities, or characteristics of the people who will interact with it.
“Nature suggests a very different approach,” Sartore and Pucci explained. “In humans and animals, the body and intelligence develop together, shaped by the environment, the tasks they perform, and the individuals they interact with. This made us wonder: What if we designed a humanoid robot by considering not only the task it needs to accomplish, but also the human partner it will work alongside?”
Optimizing Humanoid Robots and Human Interaction
Building on this concept, Sartore, Pucci, and their colleagues developed a new methodology for jointly optimizing a humanoid robot’s physical design and intelligence while taking into account the needs and characteristics of its human partners. Their framework consists of two primary stages.
The first focuses on optimizing the robot’s morphology, meaning its physical shape and structure.The team considers factors such as limb length and mass distribution across the body. Instead of optimizing these features separately, the researchers propose treating the human, robot, and task as a single interconnected system.
They applied this methodology to develop a new humanoid robot called ergoCub. The researchers specifically optimized the robot’s physical structure to reduce biomechanical strain on humans during collaborative lifting while improving its walking ability.
“After establishing the robot’s physical design, we move to the second stage: developing its physical intelligence, or the control system responsible for producing and adapting its movements,” Sartore and Pucci explained. “The robot continuously maintains an internal representation of its human partner and updates it using data collected by its sensors. This enables the robot to track the person’s movements, adjust to differences in their physical characteristics, and monitor ergonomic measures, including the biomechanical strain placed on the lower back.”
Testing the Approach With the ergoCub Robot
The researchers’ new framework builds on embodied intelligence, recognizing that a robot expresses its intelligence through its actions, behavior, sensors, actuators, physical structure, and the way these components work together.
Sartore and Pucci explained that they develop all these aspects while explicitly considering the human partner. “The robot therefore models the human partner in both its control system and physical design, embedding a form of shared intelligence directly into the robot.”
Using this methodology, Sartore, Pucci, and their colleagues developed ergoCub, a humanoid robot designed to work with people safely and comfortably. The machine is an adapted version of iCub, a humanoid robot originally developed by the RobotCub Consortium and first constructed at the Italian Institute of Technology (IIT).
After optimizing ergoCub’s embodied intelligence, the researchers conducted a series of experiments to evaluate how effectively the robot could assist a person with a physical task. In the tests, ergoCub followed a human partner while they worked together to lift and manipulate objects of varying weights.
“The robot’s name combines ‘ergonomics’ with ‘Cub,’ referring to the iCub platform that served as the foundation for the optimization process,” Sartore and Pucci explained. “When we tested ergoCub, we found that the estimated torque at the person’s lumbosacral joint was significantly lower than when the human performed the same task alone. In walking trials, ergoCub also outperformed its predecessor, demonstrating faster and more energy-efficient movement.”
Demonstrating Shared Embodied Intelligence
The study demonstrates the benefits of designing a robot’s physical structure and behavior together, while treating the robot and its human partner as a connected system. The team’s initial experiments with ergoCub suggest that this approach could make robots safer, more efficient, and more effective when performing physical tasks alongside people.
“We transformed the optimized model into a functioning humanoid robot and experimentally demonstrated that its hardware and control system work more effectively when designed as an integrated whole that explicitly considers the human partner,” Sartore and Pucci said. “We also developed and validated a method for determining the most suitable hardware parameters for a humanoid robot across multiple tasks from a whole-body perspective. This finding is significant because it demonstrates that a robot’s hardware is not simply a passive structure that software controls.”
The researchers found that designing a robot’s physical hardware and control system together allows its physical components to become an active part of its intelligence, helping it perform tasks more effectively. Their methodology could eventually optimize humanoid robots and improve their performance across a wide range of real-world applications.
“Possible applications include industrial and assistive settings where humanoid robots physically work alongside people, including manufacturing, logistics, inspection, and healthcare,” Sartore and Pucci explained. “The objective is not simply to create stronger or more autonomous robots, but to support people with physically demanding tasks while reducing biomechanical strain and workplace risks. In environments where humans and robots collaborate closely, the best design for the combined human-robot system may be very different from what would be optimal for either one independently.”
Future Research Directions
Sartore, Pucci, and their colleagues intend to further refine their methodology and apply it to the development of additional robotic systems. Their broader objective is to create robots that not only operate effectively on their own but can also achieve better results when working alongside humans.
“In our latest paper, we demonstrated that integrating a robot’s hardware, whole-body movement, and control into a unified architecture can enhance its behavior while also streamlining the development process,” the researchers said. “Instead of designing the physical body first and then relying on the control system to overcome its limitations, both aspects are developed simultaneously around the capabilities the robot needs to perform. We have already used this principle in the development of Gene.01.”
Over the past six months, the researchers have applied their methodology to optimize Gene.01, a new robot developed by GenerativeBionics. Its body, mechanical systems, movements, and sensing capabilities were developed as parts of a unified, interconnected system. So far, the robot’s performance has provided further evidence of the benefits of the team’s integrated design approach.
“Our next objective is to expand our architecture so it can systematically produce robots tailored to particular applications,” Sartore and Pucci added. “We do not expect a single humanoid design to be ideal for every task. Instead, we see Gene.01 as a versatile general-purpose platform that can serve as a foundation for specialized robots. Their artificial intelligence, sensing systems, external structure, and end effectors could then be adapted to meet the specific demands of industries such as shipbuilding, logistics, inspection, and healthcare.”
Read the original article on: Tech Xplore
Read more: Claude will now add watermarks to all content created with its tools
