Robot Identifies Objects with Thermal Sensing, 98% Accurate

image provided by pixabay

This post is also available in: עברית (Hebrew)

Robot Identifies Objects with Thermal Sensing, 98% Accurate

A new system enables robots to recognize various common and complex items using special touch sensors that detect where and when something is touched, how hot or cold it is, how rough its surface is, how hard it’s being squeezed, and its temperature.

The developing team from Tsinghua University also built a classifier with multilayered long short-term memory neural networks that work in layers to progressively identify objects based on their features, simplifying the task and improving accuracy.

While modern intelligent robots can recognize a wide range of objects using machine learning algorithms and tactile data collected by sensors, they can only do so with objects they have touched in the past. A robot gets confused upon encountering unfamiliar or similarly sized items, and its vision is further limited by background noise and objects of the same sort that differ in size and shape.

According to Interesting Engineering, the researchers aimed to imitate touch sensing in the human body by creating a robotic tactile sensing technique that included thermal sensations for more reliable and accurate item recognition. Lead author of the study Rong Zhu explained in a statement: “We propose utilizing spatiotemporal tactile sensing during hand grasping to extend the robotic function and ability to simultaneously perceive multi-attributes of the grasped object, including thermal conductivity, thermal diffusivity, surface roughness, contact pressure, and temperature.”

The humanoid robot hand is reportedly equipped with tactile sensors on fingertips and palms (mimicking human skin receptors) that capture dynamic signals during object contact, thus helping perceive both thermal and mechanical attributes at the same time.

The robot’s algorithm then “rules out object types in order, from easy to hard, starting with simple categories like empty cartons before moving on to orange peels or scraps of cloth,” stated the researchers, claiming that the approach combining the two overcomes grasping recognition complexities.

The original intent for developing this innovation was to create a robot tactile system to sort waste – the robot collected trash (like empty cartons, leftover bread, plastic bags, bottles, napkins, sponges, orange peels, etc.) and divided it into several bins for hazardous waste, food scraps, recyclables, and other waste types.

The researchers explained that their high-accuracy effective trash sorting algorithm has a wide range of applications for smart living technologies and reducing human labor. They also mention that moving forward they intend to focus on improving autonomous implementation and robotic embodied intelligence.

This study was published in ‘Applied Physics Reviews’.