Robots that can see around corners using radio signals and AI
Penn researchers developed HoloRadar, a system that reconstructs hidden 3D spaces beyond robots’ line of sight.
image:
HoloRadar uses radio waves to see around corners, allowing it to detect people at T-shaped intersections like the one pictured here.
view moreCredit: Sylvia Zhang, Penn Engineering
Penn Engineers have developed a system that lets robots see around corners using radio waves processed by AI, a capability that could improve the safety and performance of driverless cars as well as robots operating in cluttered indoor settings like warehouses and factories.
The system, called HoloRadar, enables robots to reconstruct three-dimensional scenes outside their direct line of sight, such as pedestrians rounding a corner. Unlike previous approaches to non-line-of-sight (NLOS) perception that rely on visible light, HoloRadar works reliably in darkness and under variable lighting conditions.
“Robots and autonomous vehicles need to see beyond what’s directly in front of them,” says Mingmin Zhao, Assistant Professor in Computer and Information Science (CIS) and senior author of a paper describing HoloRadar, presented at the 39th annual Conference on Neural Information Processing Systems (NeurIPS). “This capability is essential to help robots and autonomous vehicles make safer decisions in real time.”
Turning Walls Into Mirrors
At the heart of HoloRadar is a counterintuitive insight into radio waves. Compared to visible light, radio signals have much longer wavelengths, a property traditionally seen as a disadvantage for imaging because it limits resolution. Zhao’s team realized that, for peering around corners, those longer wavelengths are actually an advantage.
“Because radio waves are so much larger than the tiny surface variations in walls,” says Haowen Lai, a doctoral student in CIS and co-author of the new paper, “those surfaces effectively become mirrors that reflect radio signals in predictable ways.”
In practical terms, this means that flat surfaces like walls, floors and ceilings can bounce radio signals around corners, carrying information about hidden spaces back to a robot. HoloRadar captures these reflections and reconstructs what lies beyond direct view.
“It’s similar to how human drivers sometimes rely on mirrors stationed at blind intersections,” says Lai. “Because HoloRadar uses radio waves, the environment itself becomes full of mirrors, without actually having to change the environment.”
Designed for In-the-Wild Operations
In recent years, other researchers have demonstrated systems with similar capabilities, typically by using visible light. Those systems analyze shadows or indirect reflections, making them highly dependent on lighting conditions. Other attempts to use radio signals have relied on slow and bulky scanning equipment, limiting real-world applications.
“HoloRadar is designed to work in the kinds of environments robots actually operate in,” says Zhao. “This system is mobile, runs in real time and doesn’t depend on controlled lighting.”
HoloRadar augments the safety of autonomous robots by complementing existing sensors rather than replacing them. While autonomous vehicles already use LiDAR, a sensing system that uses lasers to detect objects in the vehicles’ direct line of sight, HoloRadar adds an additional layer of perception by revealing what those sensors cannot see, giving machines more time to react to potential hazards.
Processing Radio With AI
A single radio pulse can bounce multiple times before returning to the sensor, creating a tangled set of reflections that are difficult to untangle using traditional signal-processing methods alone.
To solve this problem, the team developed a custom AI system that combines machine learning with physics-based modeling. In the first stage, the system enhances the resolution of raw radio signals and identifies multiple “returns” corresponding to different reflection paths. In the second stage, the system uses a physics-guided model to trace those reflections backward, undoing the mirror-like effects of the environment and reconstructing the actual 3D scene.
“In some sense, the challenge is similar to walking into a room full of mirrors,” says Zitong Lan, a doctoral student in Electrical and Systems Engineering (ESE) and co-author of the paper. “You see many copies of the same object reflected in different places, and the hard part is figuring out where things really are. Our system learns how to reverse that process in a physics-grounded way.”
By explicitly modeling how radio waves bounce off surfaces, the AI can distinguish between direct and indirect reflections and determine the correct physical locations of a variety of objects, including people.
From the Lab to the Real World
The researchers tested HoloRadar on a mobile robot in real indoor environments, including hallways and building corners. In these settings, the system successfully reconstructed walls, corridors and hidden human subjects located outside the robot’s line of sight.
Future work will explore outdoor scenarios, such as intersections and urban streets, where longer distances and more dynamic conditions introduce additional challenges.
“This is an important step toward giving robots a more complete understanding of their surroundings,” says Zhao. “Our long-term goal is to enable machines to operate safely and intelligently in the dynamic and complex environments humans navigate every day.”
This research was conducted in the Wireless, Audio, Vision and Electronics for Sensing (WAVES) Lab at the University of Pennsylvania School of Engineering and Applied Science, and was supported by the University of Pennsylvania.
Method of Research
Experimental study
Subject of Research
Not applicable
Article Title
Non-Line-of-Sight 3D Reconstruction with Radar
HoloRadar relies on compact and nimble scanning equipment, opening up real-world applications.
Credit
Sylvia Zhang, Penn Engineering
HoloRadar works by reconstructing 3D scenarios from the bounces of radio waves.
Credit
WAVES Lab, Penn Engineering
Open-source modular robot for understanding evolution
A cost-effective, customizable quadruped could help researchers discover the particular advantages related to the length and segmentation of animal limbs
University of Michigan
What is it about a cheetah's build that enables it to run so fast? What gives the wolf its exceptional endurance?
While these questions can be partly answered through animal experiments, many contributing factors can't be isolated from one another. Now, a new tool has arrived: a highly customizable, open-source robot design called The Robot of Theseus, or TROT, developed at the University of Michigan.
Named in homage to Greek philosophy's "Ship of Theseus," the robot is composed of commercially available motors and 3D-printed parts, which can be rearranged to take on a broad array of designs. The plans address several pain points for animal researchers who might be able to harness robotics for biomechanical experiments, as well as for roboticists seeking more task-specific designs. Assuming access to 3D printers, the cost in parts and materials is under $4,000.
"In paleontology, we can go back and look at bones, but it's really difficult to understand how these changes in limb proportion, or in range of motion, may have affected the way an animal can move. There have been some really great insights on this question from robots that each mimic one extinct animal very precisely," said Talia Moore, assistant professor of robotics with a background in evolutionary biology and corresponding author of the study in Bionspiration and Biomimetics. "But each robot took years to design and construct.
"I wanted to make a robot that could easily shapeshift into several different extinct species proportions, so that we could compare them, and see how the evolution of those limb lengths and other features would affect their locomotion. With TROT, 60 million years of evolutionary changes in body size can happen in 20 minutes."
Usable, customizable and easy to measure
The modular robot plans and assembly guides offer three major benefits. First, they are usable by people without robotics degrees, with help from equipment that is available at many universities. As Moore pointed out, robotics offers insights into biological questions, but not many evolutionary labs have the benefit of robotics expertise.
Second, the robot's shape is highly customizable. While the published study focuses on four-legged designs, experimenters can change nearly any body segment—adding and removing parts, changing the range of motion and more. This means TROT can model most mammals and enable direct comparisons of variations on the same structure—for instance, between closely related extant and extinct species. And they can try out theoretical designs to determine whether they are biomechanically unfavorable or just untried by evolution.
Third, researchers mimicked the springiness and stiffness of muscular structures without actual springs or elastics, which can muddy measurements. TROT simulates this biological energy storage and return mechanism with backdrivable motors, which recover energy as they are driven backwards.
"Traditional robots are designed with an emphasis on industrial applications and are expensive to make. TROT was designed with ease of fabrication in mind," said Karthik Urs, a recent master's graduate in robotics and first author of the study.
"The overall part count is kept low, and most of the parts only fit together one way. That means that scientists can make most of the robot parts in-house with commodity 3D printers, assemble them and get to experimenting faster. It also makes the iteration process quick—key to enabling exploration in both robot and experimental design."
Isolating biomechanical factors that are tough to measure in animals
Moore was first inspired to make this robot when reading a 1974 experiment on running cheetahs and goats. Because the leg swings from the hip like a pendulum, physics holds that legs with more mass away from the hip, known as a greater moment of inertia, require more energy to redirect than legs that weigh the same but have most of the mass near the hip. This concept has informed the interpretation of evolutionary changes in legs—increasingly tapered limbs are likely associated with more efficient running.
However, the 1974 experiment showed that although a cheetah has a more favorable moment of inertia in its limbs, running costs nearly the same amount of energy as it does for a goat. Because so much else was different between these animals, Moore explained, the benefit from a lower moment of inertia was basically unmeasurable. In contrast, Moore's group varied only the weight distribution in their robot's limbs and was able to isolate the exact amount of energetic cost or benefit associated with that change.
TROT is designed for research and teaching rather than for operational robot work—while some 3D-printed parts break easily, they are also easy to repair and replace. Still, the results of future studies with this robot could inform commercial designs. At present, most commercial quadrupeds have fore and hind legs of the same length and style, but this test robot could reveal how to optimize the legs for the robot's intended purposes and terrains, and quantify whether the gains are worth the increase in manufacturing costs.
Researchers and enthusiasts can download the plans for the robot from U-M. The printing instructions for the parts are largely written for typical resin 3D printers, known as fused deposition modeling printers, with a stereolithography printer needed for a couple of components.
Urs is now the lead spacecraft engineer at Argo Space.
Study: The Robot of Theseus: A modular robotic testbed for legged locomotion (DOI: 10.1088/1748-3190/ae3ec1)
Journal
Bioinspiration & Biomimetics
No comments:
Post a Comment