Building an Open-Source Motorized XYZ Micro-Manipulator with Sub-Micrometer Motion Control

Diffraction Limited

Summary:

This video introduces an open-source, motorized XYZ micro-manipulator capable of sub-micrometer motion control.

• The device improves upon a manual design, offering programmable paths, higher resolution, and faster speeds.
• Key design elements include miniature ball joints, chosen for their stiffness and range of motion, preloaded with polyurethane rubber bands for stability.
• Precision motion is achieved using low-cost stepper motors driven by high-resolution PWM signals and enhanced with custom magnetic gearing for increased encoder resolution.
• The electronics are built around an RP2350 microcontroller and TB6612 H-bridge drivers, ensuring stable and jitter-free operation.
• The C++ firmware implements field-oriented control, inverse kinematics, and look-ahead motion planning, running servo loops on a dedicated CPU core.
• Calibration of the magnetic encoders corrects for nonlinearities, enabling precise control.
• Applications include automated microscopy, demonstrated by tracing a microscopic "Benchy," and optical alignment for fiber coupling.
• Future improvements include using machined metal components, ceramic bearings, and scaling up to a 6-DOF hexapod.

Overview of the Motorized Micro-Manipulator [01:18]

• The video presents a successor to a previously introduced 3D-printed manual micro-manipulator.
• The new design is fully programmable, offering:
  • Higher resolution.
  • Faster speeds.
  • A larger working volume.
• This motorized version enables a wide range of applications, some of which are demonstrated later in the video.

Miniature Ball Joints [01:52]

• Miniature ball joints are a crucial design element, offering:
  • A much larger range of motion compared to flexures.
  • High stiffness.
• The downside is non-zero friction, minimized by:
  • Choosing a small ball size relative to the connection rod length, creating large leverage.
  • Selecting suitable materials.
• The ball is fixed to the mounts, and the joint socket is placed on the connection rods.
  • This orientation prevents the ball from jumping out under large joint angles, as parallel kinematic designs primarily produce forces in the direction of the rods.
• The ball joints require preloading for bidirectional force transmission and structural integrity.
  • A small 1mm polyurethane rubber band (magic string) is used for this purpose due to its ability to maintain tension over time.
  • Each ball joint is preloaded individually and symmetrically for enhanced stability, preventing structural collapse seen with single-band preloading.

Stepper Motors for Precision Motion [03:41]

• Low-cost stepper motors are used, which can achieve high precision when driven correctly.
• Achievable step sizes are around five arcseconds in open-loop operation, equivalent to approximately 2500 times micro-stepping, using a high-resolution PWM signal.
• Mechanically, most stepper motors use ball bearings preloaded by a spring mechanism.
  • While preload is effective against play, standard ball bearings are not ideal for high-precision applications and are a potential source of error.
• The stator contains significant iron, leading to magnetic hysteresis.
  • This means the magnetic field depends on both current and previous magnetization state, making accurate open-loop positioning difficult due to hysteresis and residual friction.
  • Closed-loop control is necessary for accurate positioning.

Magnetic Encoders for Closed-Loop Control [04:53]

• A precise rotation sensor is needed for closed-loop control of the stepper motors.
• The system uses an MT6835 magnetic encoder:
  • It is affordable.
  • It offers a theoretical resolution of 21 bits (over 2 million counts per revolution).
  • Reliably measures about 100,000 counts per revolution due to noise.
• The standard use involves attaching a diametrically magnetized magnet to the motor shaft and positioning the IC behind it.
  • However, 100,000 pulses per revolution (approx. 1 micrometer step size) are insufficient for this project's resolution requirements.
• A novel magnetic gearing approach is used to increase encoder resolution without additional moving parts:
  • Bar magnets with alternating polarity are attached in an arc shape to the rotor.

    

  • The sensor is positioned tangentially with a defined vertical offset.
  • This creates a rotating magnetic field effect, as seen by the encoder chip moving along the magnet array.
  • Simulation shows consistent magnetic field rotation along horizontal paths.
  • The linearity of the magnetic field angle is good but not perfect, necessitating calibration.
• This configuration increases the overall resolution by a factor of 30.
  • The angle is unwrapped in software to provide a continuous signal.

Electronics Design [08:18]

• The control electronics are straightforward:
  • Consists of an RP2350 microcontroller (on a Raspberry Pi Pico 2 board) and three dual H-bridge motor drivers.
  • A DCDC converter supplies 5V power to logic components.
  • Encoder chips connect via a shared SPI bus with separate chip select lines.
• The PCB is double-sided with easy-to-solder through-hole components for accessibility.
  • The bottom silkscreen displays pin numbers for firmware development.
• RP2350 Microcontroller features:
  • Low-cost.
  • Hardware floating-point unit.
  • 16 high-speed PWM generators (12 used for this project).
  • Two CPU cores, allowing servo loops to run on a separate core.
• Motor drivers:
  • Typical stepper motor drivers lack the required resolution or stability (e.g., 256x micro-stepping is insufficient).
  • DRV8833 H-bridge drivers exhibited output jitter due to internal charge pumps and clocked logic, causing end-effector position fluctuations.
  • Toshiba TB6612 H-bridge driver was chosen instead:
    • It uses P-channel and N-channel MOSFETs, eliminating the need for a charge pump.
    • Contains minimal unclocked logic, providing excellent output stability.

Firmware Architecture [10:25]

• The firmware is written in C++ for simplicity, readability, and performance.
• Dual-core operation:
  • Core 2 handles servo loops for all three motors at 30 kHz for fast, responsive closed-loop control, including encoder readout, calibration, and PWM output.
  • Core 1 handles other tasks, including math-intensive inverse kinematics, command parsing, and motion planning.
• Motion planning and execution:
  • Follows standard practices: motion commands are added to a queue as path segments.
  • Look-ahead planning sets start and end velocities to comply with acceleration constraints.
  • Each path segment is subdivided into linear joint space segments.
  • Inverse kinematics is evaluated at the start and end points of these segments in a 2 kHz timer interrupt function.
  • Joint space segments are interpolated and executed, with joint positions passed to the servo loops.
• Encoder calibration:
  • The magnetic encoder setup introduces nonlinearities, requiring calibration.
  • Calibration is performed during the homing sequence, aligning the electric field with the encoder angle.
  • This method assumes the rotor perfectly follows the field, which is adequate for applications not requiring high absolute accuracy.
  • The calibration generates a lookup table (LUT) to correct raw encoder counts.
• Field-Oriented Servo Control:

  

  • Uses a cascaded control scheme: a position controller determines target velocity, which a velocity controller uses for target torque.
  • Torque command is converted to an electric field angle offset and added to the measured rotor position.
  • Without current sensing, the electric field angle directly translates to coil voltages for PWM outputs.
  • Velocity feedback is derived from the low-pass filtered derivative of the rotor position.
• Closed-loop control significantly improves precision and stiffness compared to open-loop control.

Kinematic Model [13:11]

• The firmware uses an inverse kinematic model to calculate the necessary motor angles for a desired end-effector position.
• The procedure focuses on a single arm and is identical for the other two:
  • The ball joint position on the end-effector is determined using a fixed offset from the CAD model.
  • A coordinate system is placed on the motor axis at the height of the corresponding ball joint on the actuator.
  • A circle on the XY plane represents all possible positions of the ball joint on the rotor side.
  • A sphere is centered at the end-effector's ball joint with a radius equal to the connection rod length.
  • The two intersection points between the sphere and the rotor circle are calculated.
  • The point with negative X coordinates is discarded.
  • The angle between the remaining intersection point and the X-axis yields the required motor angle.

Build & Assembly [14:33]

• This section provides a visual walkthrough of the assembly process.
• Components shown include:
  • Small ball bearings (1.5mm ball).
  • Polyurethane rubber bands (magic string).
  • Neodymium magnets (d=3mm, h=4mm).
• The assembly involves:
  • Attaching balls to mounts.
  • Securing ball joint sockets to connecting rods.
  • Integrating magnets into the rotor.
  • Assembling the motor and encoder module.
  • Connecting the arms to the end-effector.

    

Results & Applications [19:05]

Automated Microscopy [19:09]

• Test Setup:
  • Micro-manipulator.
  • Camera microscope with 395nm illumination.
  • The microscope is positioned above the end-effector, which holds the sample.
  • The setup is described as "makeshift" with 3D-printed plastic components, acknowledging potential limitations.
  • An integrated circuit die, rich in sub-micrometer structures, is used as a sample.
• Control Interface:
  • A Python program displays the live camera feed and a control panel.

    

  • It sends G-code commands via a serial interface.
  • Features include:
    • Recording and replaying waypoints.
    • Tracking a sample point to visualize motion paths and repeatability.
• Microscopy Demonstrations:
  • Step-by-step movement: The device precisely moves the sample under the microscope.
  • Complex path tracing: A 1:1000 scaled-down "Benchy" (53µm length) is traced using G-code from a 3D printing slicer.
    • A small offset is observed due to thermal drift of plastic parts heated by the camera and motors.
  • Smaller Benchy: A 20µm "Benchy" is traced at lower speeds, demonstrating finer control.
• Resolution Test:
  • Using sub-pixel accurate microscope image tracking.
  • 1 micrometer steps show excellent precision.
  • 100 nanometer steps are also precise.
  • 50 nanometer steps demonstrate the device's sub-micrometer capability, with minor fluctuations due to the non-isolated, 3D-printed setup.

    

Automated Optical Alignment [23:06]

• Application: Coupling laser light into an optical fiber, requiring precise 3D positioning (micrometer accuracy).
• Setup:
  • Micro-manipulator positions a laser source near an optical fiber.
  • A camera microscope monitors the fiber output.
  • A computer runs an optimization routine to adjust end-effector X, Y, and Z positions to maximize output brightness.
• Demonstration: The device successfully aligns the laser to the fiber, and reliably returns to the optimal position after large movements.

What's Next? [24:25]

• Areas for Improvement:
  • Replace 3D printed parts with machined metal components for better stability.
  • Improve ball joints to reduce friction by a factor of 5 to 10.
  • Swap stepper motor bearings with ceramic ball bearings for enhanced repeatability.
• Future Projects:
  • Scaling the design up to a full 6-Degrees-of-Freedom (DOF) hexapod stage.

    

• The project is open-source and available on GitHub.