This post is also available at:
Jesus (Hebrew)
Robots are increasingly being deployed in environments that are unsafe or inaccessible to humans, from collapsed buildings and underground tunnels to industrial sites and disaster zones. Many of these settings have very limited or no visibility at all. Robots often rely on cameras and computer vision to navigate, map their surroundings, and identify objects, but most vision systems are designed for visible light. When the light goes out, these systems become difficult, requiring developers to redesign the perception stack or add expensive hardware.
New machine learning approaches aim to remove that limitation. Researchers at the University of Manchester have demonstrated how a robot can operate effectively in complete darkness using a combination of infrared cameras and image reconstruction algorithms. Infrared sensors can detect reflected radiation without visible light, but their raw output often lacks the sharpness required by standard vision software. The team’s solution uses machine learning to transform thermal images into clear, camera-like images that existing robot vision algorithms can already understand.
The main advantage of this method is compatibility. Rather than rewriting navigation or object recognition software, the system restructures infrared data into a format that current algorithms can process without modification. This reduces computational overhead, reduces deployment time, and makes it easier to adapt the robot to missions with poor visibility. According to TechXplore, this approach also reduces development costs because it is built on already widely used hardware and software platforms.
In defense and homeland security applications, the implications are straightforward. Military and security robots are often required to operate at night, underground, or inside structures where lighting is unavailable or intentionally turned off. Systems that can “see” without visible light support reconnaissance, search and rescue, tunnel mapping, and inspection operations while reducing reliance on external lighting that can reveal a robot’s position. The ability to reuse existing vision software also simplifies integration into current robotic platforms.
The researchers emphasize that their work is not limited to infrared cameras. The same machine learning framework can be applied to other sensing methods such as thermal imaging and sonar, potentially extending robot perception to environments affected by smoke, dust, and other visual obstructions. This opens the door to more resilient robotic systems that can maintain situational awareness over a wider range of conditions.
Although this work is still in the research phase, it provides a practical path to improving robot perception in environments where traditional vision fails, without overhauling the systems that robots already rely on to navigate and act.
The study was published here.
