Before humanoid robots walk among us, they’ll be operated remotely, in part to gather the training data to develop Vision-Language-Action Models for autonomous, nonlethal bipedal ambulation.
There are already various approaches to teleoperation, ranging from computers linked to augmented reality accessories and game controllers to exoskeletons designed to map the wearer’s movements onto an android.
Researchers at the University of Illinois Urbana-Champaign have proposed a more familiar approach: wearing a CHILD robot in a baby carrier, where CHILD stands for Controller for Humanoid Imitation and Live Demonstration. The human operator moves the CHILD’s body parts and a larger robot learns from and imitates the movements.
“Teleoperation has become a popular approach for collecting data to solve robotic tasks in the context of Physical/Embodied AI,” explained Joohyung Kim, associate professor in electrical and computer engineering and director of KIMLAB at UIUC. “CHILD is one of the solutions we proposed to enable users to provide direct commands via teleoperation for specific target tasks.”
An academic paper [PDF] describing CHILD, by authors Noboru Myers, Obin Kwon, Sankalp Yamsani, and Joohyung Kim, is scheduled to be presented in Seoul, Korea, in October, at the IEEE RAS HUMANOIDS Conference 2025.
CHILD, Kim explained, complements a prior project called PAPRLE, which stands for Plug-And-Play Robotic Limb Environment. PAPRLE is a robotic teleoperation pipeline for connecting various input devices like puppeteers, gaming controllers, and VR-based interfaces.
“CHILD is integrated into this pipeline to support humanoid forms and dual-arm configurations, allowing portable and wireless control of target robots,” said Kim. “As a general-purpose humanoid controller, CHILD has a wide range of potential applications. It enables intuitive robot control by general users.”
Youtube Video
Kim said that during the years he spent at Disney Research, he often considered how artists could create animatronic motions more easily through more intuitive interfaces. A robot strapped to one’s chest in a baby carrier aspires to be that interface. Just wave the worn robot’s arms or legs, and its distant proxy performs the same motion.
The paper argues that CHILD has cost and portability advantages over exoskeletons and real-time motion capture systems, both of which can be exhausting because the operator is required to make all movements.
“When designing CHILD, we focused on creating an easy-to-use interface for humanoid and dual-arm robots,” said Kim.
“As mentioned in my previous posts, parents naturally teach movements to their children while using a baby carrier. In this setup, the parent’s range of motion fully encompasses the child’s, making it intuitive to scale down motions in a puppeteering manner. Most adults can use CHILD without any customization, whereas a full exoskeleton typically requires individual adjustments.”
Kim added that the compact form factor also makes CHILD lighter and generally more affordable than a full exoskeleton. The CHILD hardware specification and source code are available on GitHub.
You can 3D print the unit casing using PLA (polylactic acid) on an FDM (fused deposition modeling) printer; the entire hardware bill of materials costs just under $1,000. That doesn’t include, however, a compatible humanoid robot like the $16,000-plus Unitree G1 that would mimic the movements of the CHILD controller.
Kim contends that, in the long term, humanoid robots will become more commonplace than wheeled designs.
“Currently, wheeled or vehicle-based robots dominate the field and will continue to play an essential role in various industries,” he said. “However, if we aim to deploy more robots into intimate human environments, such as our homes, the humanoid form will likely be the only viable solution, given that our surroundings are already built for the human body.” ®