IN Brief:
- TDK and Microchip have developed a distributed sensor-node demonstration for robotics.
- The platform combines a 200MHz dsPIC33A controller, an ICM-42688-P six-axis IMU, and 10BASE-T1S Ethernet connectivity.
- Local processing supports filtering, sensor fusion, diagnostics, and machine learning before information is transmitted to a central controller.
TDK and Microchip Technology are demonstrating a distributed sensor-node platform for robotics, combining local signal processing with inertial sensing and single-pair Ethernet communication. The reference hardware uses Microchip's dsPIC33A digital signal controller alongside an InvenSense six-axis motion sensor, allowing individual nodes to process measurements before sending relevant information to a central controller.
The platform is being demonstrated at electronica 2026 in Munich. Its hardware combines a dsPIC33A controller, Microchip LAN8651/LAN8679 connectivity devices, a TDK InvenSense ICM-42688-P inertial measurement unit, and a MikroE expansion connector for compatible additional sensors. The companies say the individual components are already available, enabling developers to evaluate the architecture using established hardware.
Robotic systems can contain numerous sensing locations distributed throughout joints, actuators, grippers, and supporting structures. A humanoid robot may incorporate 100 or more sensors, according to TDK, with different devices collecting information about motion, position, force, temperature, and operating conditions. Their combined outputs must be processed and delivered to the functions responsible for movement and supervision.
Sending all measurements to a central computer can increase network traffic and the processing burden placed on that system. The amount of information generated depends on sensor count, measurement resolution, and sampling frequency. Where several sensors operate at high rates, moving their raw outputs across a shared network may become a significant part of the system architecture.
TDK and Microchip's approach places a digital signal controller at the sensing location. Each node can acquire measurements, apply filtering, combine sensor outputs, and perform selected diagnostic or machine learning functions before transmitting data. Local processing offers the possibility of reducing network traffic where the application does not require every raw sample to be forwarded.
The dsPIC33A provides a 200MHz, 32-bit processing core with digital signal processing acceleration and a double-precision floating-point unit. Its peripherals include high-speed analogue-to-digital converters and high-resolution pulse-width modulation capabilities, supporting signal acquisition and control operations alongside general embedded software.
These resources are intended for calculations that need predictable execution, such as filtering signals, combining measurements, and detecting patterns in sensor data. The demonstration includes machine learning-based gesture recognition using information from multiple sensors. The companies have not disclosed a benchmark specifying inference latency or accuracy for this particular implementation.
The ICM-42688-P supplies three-axis acceleration and three-axis angular-rate measurements. Its accelerometer detects changes associated with linear movement and the effects of gravity, while its gyroscope measures rotation. Combining their outputs can support an estimate of a robot component's movement and orientation, although measurement errors and drift must be addressed through calibration and appropriate algorithms.
Motion estimation becomes more difficult when sensors are mounted on moving assemblies. Vibration, mechanical misalignment, temperature changes, and local disturbances can affect the measurements reaching the processor. Filtering can reduce selected noise components, while sensor fusion combines information from different measurements according to their respective characteristics.
TDK identifies its PositionSense nine-axis solution as one possible expansion option. The additional sensing arrangement can support absolute orientation detection, extending the information available beyond the six-axis IMU in the reference configuration. The MikroE connector enables compatible sensor modules to be evaluated without redesigning the entire processing node.
Network communication uses 10BASE-T1S Ethernet through Microchip's LAN8651/LAN8679 technology. The standard provides 10Mbit/s communication over a single twisted pair and supports multidrop connections, allowing multiple nodes to share an Ethernet segment. This can reduce the need for separate point-to-point links between every sensing device and the main controller.
A shared communication channel must still accommodate the combined traffic generated by the connected nodes. Sampling schedules, packet transmission, network utilisation, and the timing requirements of the controlled equipment all affect the appropriate design. Local processing can reduce the volume of information transmitted, but it does not guarantee a particular end-to-end control latency.
The architecture can distribute selected computation without eliminating the central controller. Higher-level functions may still need measurements from several locations to coordinate movement, maintain stability, or detect a change in the robot's operating condition. Determining which information should be processed locally and which must be shared is therefore part of the control-system design.
Applications beyond humanoid robotics include condition monitoring of motors, pumps, gearboxes, and other industrial equipment. A node equipped with motion sensors can analyse local vibration or movement characteristics and transmit selected diagnostic information. The same processing and networking architecture can also support distributed monitoring of production machinery and automated handling systems.
Electrical integration remains important in those settings. Motor drives and switching equipment can introduce interference, while cable routing and connector reliability must accommodate moving machinery and industrial environments. Sensor placement and mechanical mounting also influence the usefulness of measurements, irrespective of the controller's processing performance.
TDK and Microchip have not published comparative figures for wiring weight, network load, or robot response time against an existing design. The electronica demonstration establishes a reference architecture using commercially available sensing, processing, and communication components. Its performance in a completed robotic system will depend on the number of connected nodes, their algorithms, and the timing requirements of the application.
