Giving a Robotic Hand a Sense of Touch: Building a Five-Finger Data Acquisition and Feedback System from Scratch

If you've played with a humanoid robot or watched related robot review videos, you must be familiar with this scenario:

A high-end robotic hand worth tens of thousands of yuan can precisely perform fixed actions like grabbing a can or crushing a paper cup, but throughout the process, it has absolutely no idea how much force its fingers are exerting. All actions are predefined open-loop commands, without perception or feedback. It's like a strong man with no sense of touch, prone to problems such as crushing objects or dropping items during grasping.

The core of a truly dexterous biomimetic hand device lies in its perception capability.

Whether it's a rehabilitation exoskeleton robot recognizing the patient's intended movement, a teleoperated robot achieving human-machine force interaction feedback, or a smart glove for gesture control, fingertip pressure data and finger joint bending angle data are the most fundamental and indispensable core signal inputs for all smart hand devices.

This project avoids expensive industry solutions: no costly ATI industrial six-axis force sensors, and no complex multi-finger array tactile skins. It builds a five-finger hand perception system from scratch using modular, low-cost, general-purpose sensors.

I. System Architecture
The entire hand five-finger acquisition and feedback system is divided into three main modules: the perception layer, the processing layer, and the feedback layer, comprehensively covering the entire process of signal acquisition, data processing, and closed-loop feedback.

(A) Perception Layer: Installing Bionic Nerve Endings on the Robotic Fingers

The perception layer is the signal source of the system. It consists of two hardware units: fingertip pressure acquisition and joint bending acquisition, corresponding to tactile perception and posture perception. The hardware selection meets the flexible wearable requirements of the human hand without restricting finger movement.

1. Fingertip Tactile Acquisition (Pressure Detection)
   Selection: 5x RP-C7.6-ST-LF Thin Film Pressure Sensors
   Working Principle: The sensors are attached to the pads of the five fingers. The force and resistance are inversely correlated. The greater the gripping pressure applied by the finger, the smaller the internal contact resistance of the sensor. The real-time force on the fingertip is reconstructed based on the change in resistance.

2. Joint Bending Acquisition (Angle Detection)
   Selection: 5 pcs 350–30AA single-bridge strain gauges (choose the appropriate length of single-bridge strain gauges based on the joint dimensions of the robotic hand fingers)
   Working Principle: The strain gauge is attached to a resilient elastic element (which can be an elastic plastic film, spring steel sheet, composite material, etc.), placed at the movable joint of each finger. When the finger bends, it drives the elastic element, stretching the sensitive grid wires of the strain gauge, causing the resistance to increase; when the finger straightens, the resistance recovers
   Key Difficulty Explanation: The resistance change caused by strain gauge deformation is only at the milliohm level, making the signal extremely weak. It cannot be directly collected by a microcontroller and requires a dedicated acquisition device for signal amplification.

(B) Processing Layer: Amplify weak signals to make hand data visually observable

Given the signal differences between the two types of sensors, a dual-collector parallel acquisition scheme is adopted. Pressure signals and angle signals are transmitted independently, paired with host computer software to achieve dynamic curve visualization.

1. Processing Fingertip Pressure Signals: CMCU-05B Multi-Channel Thin Film Pressure Collector:
2. Wiring Method: 5-way thin film pressure sensors are directly connected to the device's acquisition channels. The sensors have no positive or negative polarity distinction, making it beginner-friendly for wiring.
3. Hardware Functions: Built-in voltage divider circuit + signal amplification circuit converts the analog resistance signal into a standard voltage signal. It is equipped with a 16-bit high-precision ADC for digital quantization.
4. Local Display: The device features a 2.0-inch high-definition color screen, supports free switching between multiple units such as g/kg/N, and displays pressure values and pressure curves in real time.
5. Host Computer Communication: Supports USB/TTL serial port, communicates with the computer via the MODBUS-RTU standard protocol, enabling synchronized pressure data collection, free calibration, parameter modification, and other settings. Processing Joint Bending Signals: (Wireless single-channel one-to-five scheme or wired 5-channel scheme)

CMCU-09A Wireless Single-Channel Strain Collector, or choose the CMCU-08A Five-Channel Strain Collector.

• Wiring Mode: The strain gauge is connected to the collector using a 1/4 bridge configuration.
• Hardware Functions: Amplifies millivolt-level weak differential signals, built-in bridge algorithm converts and processes the electrical signal from the strain gauge while filtering out environmental electrical noise.
• Data Calibration: After software calibration of the original electrical signal, it is directly mapped to a standard finger bending angle of 0°-180°
• Host Computer Software Visualization  The two data streams are transmitted independently to the computer. By opening the accompanying host computer software, you can simultaneously view the pressure change curves of 5 fingertips and the bending angle change curves of 5 finger joints. Communication with microcontrollers and industrial control devices via serial port allows the robot's main controller to collect and process the data.

Feedback Layer: Build a system conditioned reflex to achieve tactile closed-loop interaction

Move beyond mere data collection by adding a hardware feedback module, achieving a complete closed loop of "Hand motion capture → System automatic judgment → Hardware feedback response".

• Trigger logic: set a fingertip pressure threshold on the acquisition device, and when the hand gripping force exceeds the threshold;
• Feedback method: the device I/O port automatically outputs a high-level signal, which can quickly light up the LED indicator with a relay;
• Expansion reserved: this implementation only achieves visual light feedback, with hardware reserved external interfaces. Subsequent upgrades can directly add vibration motors and buzzers to upgrade to real vibration tactile feedback.

II. Practical demonstration: gripping a glass
To intuitively verify the synchronous acquisition performance of the entire system, we conducted a real-world gripping test using an everyday glass. The complete test video: https://www.bilibili.com/video/BV1gx4y187dv/?spm_id_from=333.1387.upload.video_card.click

Observed phenomena during the actual test

1. Fingers naturally open: all 5 pressure curves return to zero, all joint bending angles return to zero, and the indicator light turns off;
2. Hand slowly grips the cup: all five finger joints bend simultaneously, angle curves rise synchronously; fingertips contact the cup wall and bear force, five pressure curves rise in sequence;
3. Continuous firm grip on the cup: pressure reaches the preset threshold, and the onboard LED indicator lights up synchronously;
4. Release the cup: fingertip pressure quickly drops back to zero, finger joints reset, angle curves decrease synchronously, and the indicator light turns off accordingly.

Test conclusion: the pressure data, joint angle data, and human hand movements are fully synchronized, with no delay and no data drift, accurately reproducing every subtle gripping action and force exertion of the human hand.

III. Application value

1. Lightweight wearable smart glove development for the human body
   All sensors are flexible ultra-thin devices that can be directly attached to the skin of the hand without the need for bulky mechanical shells, enabling rapid and low-cost development of wearable smart sensing gloves. Compared with commercial thousand-yuan-level smart gloves on the market, the overall hardware cost of this solution is reduced by over 70%, allowing for all-day monitoring of hand movement posture and fingertip force data, suitable for the development of wearable devices such as gesture peripheral control and human hand motion capture.

2. Tactile empowerment for bionic robotic hands
   Directly collect standard raw data of human hand gripping and touching, replicating the human force application logic and bending posture onto the bionic robotic hand. Solve industry pain points such as uncontrollable force in traditional open-loop robotic hands, crushing of fragile objects, and object detachment during grasping, filling the tactile gap in industrial robotic hands, and creating a truly perceptive dexterous robotic hand.

3. VR/AR virtual human-machine interaction
   An affordable alternative to commercial-grade motion capture gloves costing tens of thousands of yuan, it simultaneously captures dual data: finger bending posture and real fingertip pressure. It compensates for the shortcomings of traditional motion capture devices, which only capture motion without force feedback, enabling realistic touch, pinch, and grip interactions in virtual scenes, greatly enhancing the immersion of virtual reality interactions.

4. Hand Rehabilitation Medical Training
   Targeting hand dysfunction after stroke and hand trauma patients undergoing post-surgery rehabilitation, it objectively and quantitatively records two core rehabilitation metrics: finger bending amplitude and fingertip force magnitude. Using visual curve data replaces the doctor's subjective visual judgment, accurately tracking the recovery progress of hand muscles and nerves, making rehabilitation training effects quantifiable and reviewable.

5. University Maker & Embedded Teaching Training
   The project has a high degree of modularity, with the principles of pressure sensors and strain gauges being clearly independent, and the collector division of labor being clear. It is very suitable for university IoT, embedded systems, and sensor training courses. Beginners can intuitively grasp three core hardcore knowledge points: the principle of thin-film pressure sensing, strain gauge deformation angle measurement, analog signal amplification, and data visualization. Those with zero foundation can fully replicate the project.

If you've played with a humanoid robot or watched related robot review videos, you must be familiar with this scenario:
A high-end robotic hand worth tens of thousands of yuan can precisely perform fixed actions like grabbing a can or crushing a paper cup, but throughout the process, it has absolutely no idea how much force its fingers are exerting. All actions are predefined open-loop commands, without perception or feedback. It's like a strong man with no sense of touch, prone to problems such as crushing objects or dropping items during grasping.

The core of a truly dexterous biomimetic hand device lies in its perception capability.

Whether it's a rehabilitation exoskeleton robot recognizing the patient's intended movement, a teleoperated robot achieving human-machine force interaction feedback, or a smart glove for gesture control, fingertip pressure data and finger joint bending angle data are the most fundamental and indispensable core signal inputs for all smart hand devices.

This project avoids expensive industry solutions: no costly ATI industrial six-axis force sensors, and no complex multi-finger array tactile skins. It builds a five-finger hand perception system from scratch using modular, low-cost, general-purpose sensors.

🌿🌿🌿 Check our website : www.runeskee.com