ETH Zurich’s researchers, led by Michele Magno, have developed AI-based control software for their Robodog—a quadrupedal robot. The system utilizes data from built-in sensors to help Robodog learn smoother and safer movement across varied terrains. This technology aims to transform the robot into a potential aid for visually impaired individuals.
Robodog’s autonomous navigation relies on an AI-based control system referred to as “physical AI.” This system processes data from onboard sensors, including those measuring movement and spatial orientation, enabling the robot to learn smoother and safer locomotion across varied terrain. Depth cameras and a radar system further enhance Robodog’s environmental awareness, mapping surroundings and aiding obstacle avoidance even at higher speeds. Researchers are also testing new sensors to increase Robodog’s independence, demonstrated by its success in a robotics competition where it flawlessly completed a challenging obstacle course ten times consecutively.
The robot’s adaptability extends to potential attachments like a robotic arm, controlled by sensors in specialized glasses that respond to the wearer’s gaze, allowing it to assist individuals with physical impairments by performing tasks like opening doors or retrieving objects.
Myosuit: Adaptive Exoskeleton Supports Lower Body Rehabilitation
The Myosuit is a novel exoskeleton designed to aid individuals with lower body paralysis, offering a flexible and lightweight alternative to traditional metal systems. Combining textile elements with plastic components and small motors, the suit assists hip and knee joints, reducing the effort needed to overcome gravity during walking. Importantly, built-in sensors monitor the user’s exerted force, allowing the Myosuit to adapt and provide precisely tailored support. This self-learning device isn’t limited to clinical settings; it can be worn comfortably over everyday clothes, enabling patients to practice physiotherapy at home and regain mobility more effectively.
Recent studies, including a collaboration with the Charité University Hospital in Berlin, demonstrate its potential to help patients overcome limitations from conditions like stroke, spinal cord injuries, muscular dystrophy, or even heart failure – one patient increased their walking distance from 100 meters to several kilometers, even completing portions of the Zurich Marathon.
To use it for our purposes,” he says, “we first need to modify it.”
Michele Magno
“Physical AI” Integrates Sensor Data for Smoother Robotic Movement
Researchers are developing “physical AI” for robots like Robodog by utilizing data from built-in sensors. Depth cameras and radar systems are also integrated, allowing Robodog to detect distances to objects and map its surroundings in greater detail, even without external input. This AI-driven approach extends beyond navigation; the system adapts to the user’s capabilities in other robotic devices. For example, the Myosuit exoskeleton uses sensors to measure a wearer’s exerted force during walking, learning to provide the precise amount of support needed for hip and knee joints—facilitating both rehabilitation and increased mobility. This sensor integration is vital for creating robots that can effectively assist individuals with physical impairments or aid in recovery from conditions like stroke or paralysis.
Magnetically Guided Bacteria Target Cancer Drug Delivery
Researchers are also investigating using magnetically guided bacteria for targeted cancer drug delivery. These bacteria, specifically E. coli, are coated with approximately one thousand iron-oxide nanoparticles, enabling external magnetic fields to direct them through the bloodstream. This approach aims to improve drug efficacy by concentrating the therapeutic agent directly at the tumor, minimizing side effects from widespread drug distribution. A key challenge in cancer treatment is ensuring sufficient drug concentration reaches the tumor site, often causing clinical trial failures. Utilizing bacteria as carriers offers a potential solution by delivering agents directly to the targeted area and reducing exposure to healthy tissues. This “physical AI” approach, while microscopic, represents another innovative robotic application being developed at ETH Zurich for medical advancements.
