🤖 Melexis's Haptic Fingertip Module with True 3D Force Sensing!

[Cengiz Özemli]

Introduction: A New Era in Robotic Grasping​

Melexis has launched its new MLX90901 tactile fingertip module, set to revolutionize the grasping capabilities of humanoid robots. This module combines Tactaxis magnetic technology with complete mechatronic integration, helping robot developers simplify prototyping, enhance grasping skills, and accelerate physical AI.

Current Challenges and Melexis's Solution​

While today's autonomous systems can plan complex tasks, limited sensory feedback at the point of physical contact poses a barrier to implementation. Standard industrial grippers often lack tactile sensors or rely on single-axis force measurements that detect pressure normal to the surface but cannot register lateral slip. Although multi-axis tactile sensors exist in academic laboratories, they often require specialized elastomer formulations, complex wiring, and manual calibration, hindering high-volume production.

Melexis's MLX90901 module is designed to address these issues. To manipulate irregular or fragile objects without dropping or crushing them, robotic control loops need real-time data on both the magnitude and direction of contact forces. Resolving forces along three orthogonal axes allows a robotic hand controller to detect slip and dynamically adjust grip strength.

Operating Principle and Mechatronic Integration​

The MLX90901 resolves three-dimensional contact forces by combining Melexis's magnetic Tactaxis sensing technology with a complete mechanical and electronic assembly. In a magnetic tactile sensor, external mechanical loads deform a compliant elastomer layer embedded with a permanent magnet. A multi-axis Hall effect magnetic sensor beneath the elastomer measures the resulting changes in the magnetic flux vector, converting these variations into normal and shear force components.

Since the performance of a magnetic tactile sensor heavily depends on the mechanical properties of the elastomer and the precise alignment of the magnet, integrating bare sensor dies incurs significant engineering overhead. Melexis solves this problem by packaging the sensor die, permanent magnet, compliant elastomer, structural housing, and signal conditioning electronics into a standardized fingertip form factor. This combined architecture eliminates the need for robotic hand manufacturers to source and calibrate separate mechanical and magnetic subsystems.

Measurable Performance and System Specifications​

The MLX90901 is designed to capture subtle force variations during delicate handling and withstand standard manipulation loads. Key performance parameters include:

• Normal force measurement range: 15 N
  
• Shear force measurement range along two orthogonal axes: 4 N
  
• Full-scale measurement accuracy: 10%
  
• Typical force resolution: 200 mN
  
• Digital communication via I2C interface for direct connection to embedded hand controllers
  

According to Julien Ghaye, Product Line Manager at Melexis, designing the complete mechatronic module rather than providing a standalone sensing element allows the company to manage the interdependent material, mechanical, and electronic variables essential for high-volume production.

Development Workflow and Prototyping Support​

To support system integration, Melexis also provides the DVK90901 evaluation kit alongside the fingertip module. Robotic hand manufacturers and OEMs can use the kit to characterize the sensor's three-axis force response, benchmark performance against specific manipulation requirements, and collect contact datasets to train closed-loop control algorithms.

This release follows four years of development, during which Melexis first demonstrated the biomimetic Tactaxis principle using discrete tactile pixels or taxels. Subsequent customer trials informed the transition from individual sensor components to a standardized, production-ready fingertip module suitable for humanoid robotic platforms.

Additional Information: Transition to Physical AI​

The development of integrated three-axis tactile modules reflects a broader shift in robotics toward physical AI, where sensorimotor data is used to train foundational models for manipulation. Historically, commercial tactile sensing has been divided among capacitive arrays, piezoresistive films, and optical-tactile sensors like GelSight-style cameras. Optical tactile sensors offer high spatial resolution by monitoring the deformation of an internal gel layer with a miniature camera, but they require significant internal volume, higher power consumption, and computationally intensive image processing.

Magnetic tactile sensors like the MLX90901 offer an alternative architecture characterized by a compact physical footprint, low power consumption, and low-latency digital output over standard serial buses like I2C. By outputting three-axis force vectors directly instead of raw image frames, magnetic modules reduce the computational load on local joint controllers. However, magnetic sensing architectures must manage external magnetic interference, which is typically addressed through differential sensing geometries or localized magnetic shielding within the sensor package.