Scientists Develop Robotic Arm That Can Feel Touch, Recognize Tickling, and React Like a Human Arm

A robotic arm that can detect contact across its own structure and change its movements accordingly could help address a persistent weakness in robotics: responding to physical interactions that cameras cannot reliably see.

The system, called EmArm, was developed by a team at the Hong Kong University of Science and Technology (HKUST). It combines a rigid mechanical skeleton with a soft optical tactile layer, allowing it to interpret touch, stabilise objects, and alter its route when an obstacle enters its path.

Rather than treating touch as a simple warning signal, the design uses tactile information to help guide the robot's next action. In experiments, it recognised five types of human contact with reported accuracy of up to 99.8%, while separate demonstrations showed it balancing an off-centre basketball and responding to obstacles hidden from view. The findings were published in Nature Sensors on 20 July 2026.

A Tactile Skin Turns Contact Into Data

EmArm's tactile layer covers more than 71,000 square millimetres across its upper and lower arms. Tiny cameras inside the structure analyse how forces are distributed across the skin, turning physical contact into information that the control system can use.

The system has a reported spatial resolution of 0.085 millimetres. Its two-point discrimination threshold can reach 18.2 millimetres, meaning it can distinguish two nearby contact points at a spacing finer than the typical 30–50 millimetre range reported for human arm skin.

The optical approach also reduces the need for complex wiring associated with electronic skins. During testing, the system maintained sensing fidelity of up to 97.9% across five types of environmental interference, including electromagnetic, light, temperature, humidity, and acoustic disturbances.

These measurements matter because contact with an object is rarely uniform. A robot needs to determine not just whether it has touched something, but where the contact occurred, how much force was applied, and whether that information requires a change in movement.

Recognising Tickling Without Feeling Emotion

EmArm was tested on five touch categories: stroking, pinching, poking, tickling, and hitting. It can map those patterns to feedback expressions associated with happiness, sadness, curiosity, shyness or anger.

The reported recognition accuracy of up to 99.8% applies to the tested interaction-recognition task. It does not mean the machine experiences tickling or feels emotions as a person would. Instead, it identifies patterns in tactile signals and uses them to select a corresponding programmed response.

That process combines tactile data with proprioception, the system's information about its own joint positions. By interpreting both sources together, EmArm can connect what is happening at its surface with the position of its mechanical parts and feed the resulting information into its motion-control system.

The approach offers a different way to design human–robot interaction: physical contact becomes information that can shape the machine's response, rather than an event that necessarily interrupts a task.

Touch Helps the Robot Adapt When Vision Fails

One demonstration involved keeping a basketball stable when it rested away from the arm's centre of balance. EmArm used tactile feedback and force control to adjust its posture, helping prevent the object from slipping out of position.

In another test, a dynamic obstacle, such as a human hand, entered the robot's path where its vision system could not detect it. The arm used touch to detect the obstruction, stop, assess whether it was temporary or persistent, and replan its route.

The system also retained repeat-positioning accuracy comparable to conventional robotic arms. HKUST reported that it withstood loads exceeding 15 kilograms and torsional loads of 7.35 newton-metres, although these figures describe separate mechanical tests rather than a guarantee of performance in every setting.

The technology remains a research system, not proof that robots are ready for unrestricted use around people. Its broader significance is that tactile sensing is built into the arm itself and linked directly to movement decisions.

If the approach can be extended and validated beyond laboratory demonstrations, it could inform future robots designed to handle objects, navigate cluttered workspaces or interact more adaptively with people in industrial, domestic, and assistive settings.