04 Power Drive: The 3-Phase MOS Bridge Determines the Joint's Output Torque Capacity

Figure 5: The three-phase motor wires are centrally connected to the MOS power section, with sampling resistors, drivers and filter components arranged around it.

 

This joint driver is designed for three-phase brushless motors. High-performance control of three-phase BLDC motors typically requires three half-bridges, forming a three-phase inverter with a minimum of six power MOSFETs. Multiple large-package MOS devices can be spotted on the PCBA surrounding the three-phase output solder points, consistent with this architecture.

 

The operating logic of the three-phase bridge can be simplified as follows:

Battery or bus voltage is fed in via P+ and P-; High-speed switching of MOS half-bridges converts DC voltage into three-phase current;

The three-phase current flows into motor windings to generate a rotating magnetic field; The controller continuously adjusts PWM duty cycles based on encoder position and current feedback;

Ultimately enabling the joint to output target torque, speed or position.

Poor design of this section will lead to several typical issues: excessive MOS heating, low-speed jitter, severe torque ripple, false overcurrent protection triggering, and EMI interference affecting encoders or communication circuits. For this reason, power board design for robot joints is far more challenging than that of ordinary small motor driver boards.

 

Figure 6: The TLV767 voltage regulator can be seen on the board, alongside inductors, capacitors and several low-voltage power components.

The power supply tree is also worth analyzing. LDO regulators such as the TLV767 and power inductors marked "100" are visible on the PCB. Considering the application scenario of 48V-rated joints, it can be inferred that the low-voltage system is unlikely to be directly stepped down from high voltage solely via LDOs. Instead, a power supply architecture similar to the following is adopted:

High-voltage bus → Switching step-down conversion → Low-voltage rails (5V / 3.3V, etc.) → Secondary LDO regulation / local noise filtering

This power architecture is more rational. The MCU, encoder, sampling amplifiers and communication transceivers require clean, stable low-voltage power, while the MOS switches and motor phase lines generate intense noise. If the low-voltage power supply lacks sufficient anti-interference performance, the first faults to emerge are usually not motor malfunctions, but abrupt angle jumps, communication dropouts or drifting current sampling readings.

05 Current & Temperature Sensing: Robot Joints Must Monitor Output Torque in Real Time

 

Figure 7: Multiple R001 low-value sense resistors are located beneath the three-phase power section.

The R001 components are critical on this board, representing milliohm-level low-value sense resistors (1 mΩ typically) used to measure motor phase current or bus current. Current sensing forms the core of closed-loop control for robot joints, as the output torque of a BLDC motor is strongly correlated with its phase current.

In other words, the controller does not "guess" the torque output of the joint; instead, it calculates the current by measuring the tiny voltage drop across the sense resistors:

Voltage drop across sense resistor → Amplification & filtering → ADC sampling → Current calculation → FOC torque control

The precision of this signal chain directly affects:

 Torque Control Accuracy: If the current estimation is inaccurate, the torque output will be inaccurate.

Low-Speed Smoothness: High sampling noise tends to cause jitter at low speeds.

Overcurrent Protection: Slow response may burn out MOSFETs, while overly sensitive settings result in false protection triggers.

Thermal Management Strategy: Sustained high current leads to simultaneous temperature rise in motor windings, MOSFETs and current-sense resistors.

From a layout perspective, placing current-sense resistors close to MOSFETs and phase wiring shortens the high-current loop, which helps reduce parasitic inductance and sampling errors. A large number of small resistors and capacitors are arranged near each phase, generally used for gate damping, sampling filtering, driver stabilization and local decoupling.

Temperature detection is equally indispensable. The MOTOR NTC label indicates a thermistor installed inside the motor or adjacent to the windings. For humanoid robot joints, low-speed high-torque operation, prolonged static holding, and transient impact during standing up or squatting all generate substantial temperature rises. The NTC enables the controller to perform three key temperature-based functions:

Current limiting: Reduce the maximum output current as temperature rises; Derating: Cap peak torque under sustained heavy loads;

Protective shutdown: Enter fault mode once temperature exceeds the safety threshold.

This is also a fundamental capability that professional robot joints must have. Without temperature closed-loop control, the joint cannot operate reliably for extended periods despite its high peak torque rating.

06 Position Detection: Dual Magnetic Encoders and Gears Convert Single-Turn Angles to Multi-Turn Position Readings

After removing the PCBA, a set of small gear structures can be seen on the reverse side. These are not the main reduction gears, but auxiliary mechanisms designed for angle detection.

 

Figure 8: Gear and magnetic encoder area on the reverse side of the control board

 

Figure 9: Rotation of the central shaft drives the two adjacent pinion gears to rotate synchronously.

Structurally, rotation of the central shaft drives the two small gears. A magnetic component is fitted at the center of each small gear, with magnetic encoder ICs mounted on the matching positions on the rear side of the PCBA. Magnetic encoders measure angular variations of magnetic fields, eliminating the need for optical grating discs. They are tolerant to minor oil contamination and dust, making them more suitable for fully enclosed robot joints.

 

Figure 10: The dual-gear and dual-encoder structure improves angular resolution and enables multi-turn position detection.

 What stands out most here is the "dual-gear + dual magnetic encoder" configuration. Its value goes far beyond simply adding a second sensor for higher accuracy; it is most likely engineered to realize a multi-turn absolute position encoding scheme:

 A single magnetic encoder accurately measures angles within 0–360°;

yet a joint may rotate multiple revolutions, and a single-turn encoder cannot determine the number of rotations on its own;

If the two gears feature different tooth counts or gear ratios, they form a vernier-like correlation;

By combining the two angular readings, the controller calculates the absolute position over a much wider range.

 This structure delivers great value to humanoid robots. After power-off, the legs, arms or waist joints may be displaced by external forces. Upon re-powering, joints requiring extensive homing movements suffer from low efficiency and pose safety hazards. Multi-turn absolute position sensing allows joints to instantly retrieve posture data as soon as power is restored.