On August 6, Zhidongxi reported that a research team from Beijing University of Aeronautics and Astronautics has recently developed a new tactile sensing technology that enables robots to have human-like tactile senses, such as feeling the texture of an object's surface and how hard it is.
This technology can be applied to robots made of soft materials, facilitating the creation of bionic prostheses and humanoid robots. Researchers say the study was inspired by the proprioceptive mechanisms of humans and mammals, a biological mechanism that allows them to perceive or be aware of their own position and movement.
The research team conducted a series of tests to evaluate the tactile sensing technology of their prototype system and found that the system achieved 100% accuracy in object texture recognition and 99.7% accuracy in stiffness recognition.
The research has been preprinted on arXiv, with the title "Tactile Sensing with a Tendon-Driven Soft Robotic Finger".
I. Researchers draw inspiration from the human body to give robotic hands the sense of touch.
In recent years, many robotics experts around the world have been trying to develop a robotic system that can replicate human touch, while also trying to use soft materials to replace rigid structures to create more realistic and advanced bionic limbs and humanoid robots.
Soft materials have an advantage in texture acquisition for collecting tactile information, but robotic arms made of soft materials often cannot collect a wider range of sensory information. To date, replicating the biological mechanisms by which humans collect tactile information about objects remains extremely challenging.
Past research has often used rigid fingers to measure contact stiffness. These methods estimate the contact force between the finger and the object and the finger's displacement based on the finger's kinematic model to calculate the contact stiffness. However, this method is difficult to apply to soft robot fingers because soft robotic hands are too flexible, kinematic modeling is difficult, and the results are usually not very accurate.
Inspired by the proprioception framework found in humans, the research team at Beijing University of Aeronautics and Astronautics developed a new tactile sensing technology that gives soft robotic hands the ability to sense touch and perceive the texture and stiffness of objects.
“The ability to still feel the posture of your hands, the position of your arms, or the weight of a bag of groceries when you are blindfolded and your ears are covered is called proprioception. We have been working on a prosthetic hand research project and trying to find ways to give prosthetic hands tactile feedback,” said Chang Cheng, one of the participants in the study.
II. Tactile machines can achieve 100% accuracy in recognizing finger textures.
In the past, robotics researchers typically did not associate proprioception with touch because human proprioceptive mechanisms do not provide particularly precise feedback. However, industrial sensors are far more sensitive than human proprioceptive organs, and using them in robotic fingers can help researchers collect more accurate tactile feedback.
The robotic hand prototype system created by the research team at Beihang University consists of a linear actuator, tendon/cable, strain sensor, and a soft robotic finger. In addition, a polyurethane sleeve is placed over the end of the finger to mimic the human fingertip.
The tendons of the robotic finger are connected to the actuator, and strain sensors are installed in the middle of the tendons. When the actuator is driven, it pulls the tendons to bend or straighten the finger, and the strain on the tendons changes accordingly. When the finger touches different objects, the sensors output a series of strain signals to characterize the touched objects.
Researchers experimented with the robotic finger using eight different textured plates and four cylinders with different stiffnesses.
In the tactile sensing test, researchers allowed a mechanical finger to glide slowly across the surface of a textured plate, and the resulting tendon strain was recorded. A total of 60 tests were conducted on each textured plate.
In the stiffness sensing test, a cylinder used for the experiment is placed under a finger. Once the finger is activated, pressure is applied to the object, and this state is maintained for 4 seconds. Data on tendon strain at each stage of the experiment are also recorded.
The research team then used machine learning models to analyze this data, ultimately identifying the corresponding object textures or object stiffness. Verification showed that this method achieved 100% accuracy in object texture recognition and 99.7% accuracy in object stiffness recognition.
Third, this technology will be used to develop robots and prosthetics.
“Most existing research on neurally controlled bionic fingers involves mounting sensors on the surface of the fingertip. While these studies have yielded good results, they require precise contact between the fingertip sensor and the object, which is often not guaranteed in practice,” Chang Cheng said. “The key to our research is that the sensing unit is located on the tendon, so any contact on the finger can generate a characteristic signal output, which can be used to infer tactile information.”
The research team embedded sensors into the tendons of robots to achieve a new tactile sensing method, which is unprecedented and has great application potential after testing.
In the future, this system they developed could be used to develop more advanced robots and prosthetics. These robots and prosthetics could have tactile feedback and utilize proprioceptive feedback without requiring perfect or precise contact with object surfaces.
Chang Cheng said, “We are now exploring the system’s slip detection capabilities. Slipping is inevitable when humans manipulate or grasp something, so detecting and controlling slippage is crucial for a robot to be able to control objects robustly and reliably. We believe slip detection will be a great feature to add, and our initial experiments have shown very promising results.”
In addition to further developing this system, the research team is also collaborating with a nanotechnology lab to develop a low-cost tactile sensor that can detect force or torque signals and can be placed on a robot's fingertips. They have already created a prototype of this device and are evaluating its performance.
Conclusion: Tactile sensing technology enables prosthetics to also have the sense of touch.
Soft materials have a significant advantage in collecting tactile information, and due to their softness, they are more suitable for manufacturing bionic limbs and humanoid robots compared to rigid materials.
In the past, it was quite difficult to give soft robots a sense of touch. However, the research team drew inspiration from humans and mammals and cleverly solved this problem, enabling the robot's fingers to not only recognize the texture of objects but also their stiffness.
This technology will be used in robot development and prosthetic manufacturing, enabling robots and prosthetics to also have tactile sensation.
