一.Introduction to the humanoid robot DC power supply system

A humanoid robot is a complex system integrating mechanics, electronics, computing, communication, and sensing. Its powerful motion capabilities and intelligent interaction performance depend entirely on a stable, efficient, safe, and lightweight internal power supply network. This network acts like the circulatory system of the human body, continuously delivering energy to the robot's brain (main control chip), muscles (joint drivers), and senses (various sensors). A well-designed DC power supply system must not only meet the dynamic demands of high peak power (such as jumping and running) but also achieve the highest energy conversion efficiency within limited volume and weight constraints to extend runtime. At the same time, it must have comprehensive electrical protection and electromagnetic compatibility (EMC) design to ensure stable operation in complex electromagnetic environments and safeguard human-robot collaboration safety.

Earlygeneration and smallsize robots often used 24V or even lower voltage platforms, but for humanoid robots requiring high torque output, a higher voltage platform is an inevitable choice. The core logic is that according to the power formula P = V × I, when outputting the same power (P), increasing the system voltage (V) can effectively reduce the operating current (I). Lower current brings two core advantages:

1. Reduced line losses: The power loss in cables is proportional to the square of the current (P_loss = I² × R). Halving the current theoretically reduces line losses to one-quarter of the original, directly improving energy transmission efficiency from the battery to the load—crucial for extending runtime.

2. Reduced harness weight: Lower current allows the use of thinner wires (smaller wire gauge), significantly reducing the weight and volume of the entire robot's internal power harness. For robot designs that count every gram, this advantage is critical.

At present, the 48V DC bus has become the industry-recognized mainstream standard. The main basis for this choice is the Safety Extra-Low Voltage (SELV/ES1) standard defined by the International Electrotechnical Commission (IEC 62368-1 / formerly IEC 60950-1 for IT/AV equipment), whose maximum DC voltage limit is typically 60V. The 48V platform provides high power capability while remaining at the upper limit of the safety voltage range, balancing performance and safety without requiring expensive and complex high-voltage insulation protection measures.

When a humanoid-robot walks, runs, carries objects, changes posture, or is subjected to external disturbances, the joint drive motors face severe transient load fluctuations—rapid jumps or drops in torque over short periods cause bus current spikes several times the steady-state value, severe bus voltage fluctuations, and even system electromagnetic interference (EMI) exceeding limits, power device overcurrent breakdown, and driver undervoltage lockout faults. To solve this core technical challenge, mainstream humanoid robot manufacturers are researching various technical routes and attempting measures at multiple levels to systematically offset the impact of transient loads.

Diagram of a humanoid robot and its DC power supply system

二. Causes and hazards of transient loads in humanoid- robot DC power supply systems

1 Core causes of transient loads

The joint actuator unit of a humanoid-robot is essentially an integrated servo drive system. The frameless torque motor is the mainstream power output component—from a technical principle perspective, the output torque of the motor is proportional to the phase current. When the mechanical load on the joint changes suddenly, the motor's electrical power demand jumps within an extremely short time (typically milliseconds or even microseconds), which is the root cause of fluctuations in the power supply chain. From the perspective of actual operating conditions, typical scenarios that trigger transient loads include three categories:

· Startup/stall current spikes: At the instant a joint motor starts from a standstill, due to mechanical inertia, the motor's output torque demand can reach 2-5 times the steady-state value in a short period, and the corresponding input current jumps synchronously. Taking a certain brand's Hx model robot knee joint actuator as an example, its output torque limit is 360 N·m, and the instantaneous current peak during startup can reach more than 3 times the steady-state value.

· Dynamic load mutations: When the robot is running at high speed, crossing obstacles, carrying heavy objects, or subjected to external impact disturbances, the joint motor's output torque must be adjusted significantly within an extremely short time to maintain posture stability. Under such conditions, the rate of load change is typically much higher than the motor's electrical response speed, directly causing severe fluctuations in bus current.

· Regenerative energy surges: When a joint motor is in a braking state or is driven in reverse by a load (such as the knee joint motor during a squatting phase), the motor's operating state switches from motoring mode to generating mode, converting mechanical energy into electrical energy and feeding it back to the DC bus. If this regenerative energy cannot be consumed or recovered in time, it directly causes voltage spikes on the bus, which in severe cases exceed the withstand voltage threshold of the power devices.

2 Fatal hazards to the power supply and drive system

The transient load of joint drives is not merely a single current fluctuation problem; it propagates backward along the power supply chain, triggering a cascading risk of "current spike – voltage fluctuation – electromagnetic interference – device failure." Specifically, the hazard propagation path includes four layers:

· Overcurrent/overvoltage breakdown risk of power devices: The peak current corresponding to transient loads generates back EMF across the parasitic inductance of the power supply line, causing instantaneous bus voltage dips. When regenerative energy is fed back to the bus, it causes instantaneous bus voltage rise. Taking the industry-mainstream 48V power bus as an example, the amplitude of such voltage fluctuations can exceed ±20%. Without targeted protection design, whether it is driver latch-up and restart caused by undervoltage, or direct breakdown of power switching devices caused by overvoltage, both will result in joint drive failure.

· Stability of the power supply chain is compromised: The multi-joint drive system of a humanoid robot adopts a common DC bus architecture—all joint drivers are connected in parallel on the same bus, sharing the energy input. Under the robot's dynamic motion conditions, the transient load currents of multiple joints superimpose on the bus, causing a significant increase in bus voltage ripple and even harmonic resonance. This fluctuation is conducted through the power supply line to other joint drivers operating in steady state, causing reference point offset in the current loop control, ultimately resulting in joint output torque fluctuations and reduced motion precision, and in severe cases causing the robot to lose balance and fall.

· Electromagnetic interference (EMI) exceeding limits, causing control logic faults: The high rate of change of peak current (di/dt) generates high-frequency noise on the parasitic inductance of the power loop. This noise radiates outward through cables or is conducted to other sensitive circuits through power lines. The control circuits of joint drives include weak-signal circuits such as encoders, current sampling, and communication buses, which have poor anti-interference capability. If the EMI intensity exceeds the circuit's immunity threshold, it causes sampling data distortion, communication packet loss, and motor rotor position detection deviation, ultimately leading to confusion in motor control logic.

· Accelerated lifespan degradation of the energy unit: If a single lithium battery is used for power supply, the high-rate discharge current of transient loads generates a large amount of heat on the battery's internal resistance, causing a rapid rise in internal battery temperature. The reverse feeding of regenerative energy causes the battery's charging current to be momentarily excessive. Both conditions accelerate the aging of the lithium battery's internal electrode materials and significantly degrade cycle life. According to industry measurement data, without a supercapacitor module, the cycle life of a lithium battery degrades from the conventional 2,000+ cycles to less than 1,000 cycles.

三. Technical measures and solutions for humanoid- robot DC power supply systems to cope with transient loads

For the multi-dimensional hazards of transient load impacts, a single technology cannot be used to address them. Instead, a four-layer collaborative optimization architecture of "power source layer – distribution network layer – power conversion layer – control layer" must be built, targeting the impact of transient loads from four dimensions: energy reserve, impedance matching, noise suppression, and dynamic response. The core design logic of this architecture is "layered compensation, graded filtering, and coordinated control"—supplementing transient energy reserves on the power source side, optimizing path impedance on the distribution side, rapidly limiting current fluctuations on the drive side, and coordinating the allocation of instantaneous power demand at the system level.

1 Power source architecture: hybrid energy storage for transient power compensation

This is the most fundamental and critical link in solving transient load impacts—its core logic is to use the complementary characteristics of different energy storage elements to separately address continuous power and instantaneous high-power demands. Specifically, the lithium battery is responsible for providing continuous power for conventional scenarios such as walking and working, while the supercapacitor serves as a peak power compensation unit, specifically responsible for rapidly releasing instantaneous power during load mutations to smooth out the battery's power fluctuations. The complementary characteristics of the two exactly cover the full operating range of power demands of a humanoid robot.

The mainstream topology architecture in the industry is the "semi-active DC/DC hybrid energy-storage architecture," which is adopted by the power compartment of a certain brand's Hx model robot. The design details of this architecture are that the main energy bus uses a 48V high-rate lithium battery pack, connected in parallel with a supercapacitor module controlled by a bidirectional DC/DC converter. The key here is that the supercapacitor is not simply connected in parallel with the lithium battery, but achieves power decoupling from the lithium battery through a bidirectional DC/DC converter—meaning that the bidirectional DC/DC converter can dynamically adjust the charge/discharge state of the supercapacitor in real time, as well as the magnitude of power compensation to the bus, so that the lithium battery only needs to provide continuous steady-state power, and the transient power demand is entirely borne by the faster-responding supercapacitor.

This solution has three technical advantages:

· Fast transient response: The power density of supercapacitors is several times that of lithium batteries, with a response speed reaching the millisecond level—at the instant of load mutation, the supercapacitor can release a large current first before the lithium battery's output current changes significantly, rapidly compensating for the power gap and controlling the bus voltage fluctuation amplitude within a reasonable range.

· High energy recovery efficiency: When the joint motor is in a regenerative braking state, the supercapacitor can preferentially recover this feedback energy, rather than allowing this reverse current to directly impact the lithium battery. When the system subsequently experiences a power gap again, the supercapacitor can release this stored energy back to the bus, reducing the instantaneous charge/discharge pressure on the battery and improving the overall system energy efficiency, with measured efficiency improvements of up to 8%.

· Extended battery life: The "peak shaving and valley filling" effect of the supercapacitor limits the lithium battery's operating current change rate to a low level, significantly reducing the battery's dynamic load stress. According to industry measured data, after pairing with a supercapacitor module, the cycle life of the lithium battery can be increased from less than 1,000 cycles to around 3,500 cycles.

2 Distribution network architecture: common bus + graded filtering to reduce path impedance

The distribution network is the intermediate link connecting the energy side and the load side. The voltage fluctuations caused by transient loads are essentially the voltage drop of current across the distribution line impedance. Therefore, the design goal of the distribution network is to reduce path impedance as much as possible while filtering out ripple in different frequency bands in stages, ensuring low impedance and high stability of power transmission. The mainstream distribution network design for humanoid robots adopts a combined architecture of "common DC bus + graded capacitor energy storage + π-type EMI filtering."

（1） Common DC bus design

Multiple joint drivers sharing a DC bus is a mature technical solution for industrial multi-axis servo systems and is also the optimal choice for humanoid robots. The core advantage of this solution is that it enables in-situ balancing of regenerative energy within the system—when some joint motors are in generating/braking state, the regenerative energy they produce can be directly supplied through the bus to other joint motors in motoring state, without the need for secondary conversion through energy storage units, resulting in higher energy recovery efficiency.

The technical details of this solution are: the DC sides of all joint drivers are connected in parallel on the same bus. The bus uses high-current busbars or multi-strand copper-core shielded cables to minimize line impedance as much as possible. Taking the 48V power bus of a certain brand's Ox model robot as an example, its bus current-carrying capacity is designed strictly to more than 3 times the sum of all joint steady-state currents, and the ratio of wiring width to copper thickness fully complies with high-current carrying design specifications. The goal is to control the line voltage drop under transient high currents to within 5% of the bus voltage.

（2） Graded capacitor energy storage and filtering

To further compensate transient power, reduce bus impedance, and filter noise in different frequency bands, capacitor energy storage and filtering units must be placed on the DC bus—this is the core hardware means of suppressing voltage fluctuations. In engineering, a combined design of "multi-stage capacitor array + π-type EMI filtering" is adopted, arranging capacitors in stages along the power transmission path according to their different characteristics to separately handle ripple in different frequency bands.

The topological details of this solution are: from the energy side to the load side, a first-stage energy storage capacitor bank, a π-type EMI filter unit, and a second-stage high-frequency decoupling capacitor group are arranged in sequence. In terms of parameter selection, the graded capacitor values must be matched according to the bus voltage and actual peak current:

· First-stage energy storage capacitor bank: Arranged at the input end of the bus, near the output terminals of the supercapacitor module and lithium battery pack. It is mainly used to store and release relatively large amounts of energy, compensating for power gaps in the medium and low frequency bands. Since the output terminals of lithium batteries and supercapacitors often have some parasitic inductance, under high-frequency conditions this inductance impedes rapid current changes; the first-stage energy storage capacitor bank can provide this energy locally when transient current demand arises, offsetting the effects of parasitic inductance. In terms of voltage rating, considering transient spikes on the bus, a 48V bus system must select capacitors with a voltage rating of 80V or above, and a 60V bus system must select capacitors with a voltage rating of 100V or above, reserving sufficient safety margin. In terms of capacitance, it must be matched according to the joint's peak current: when a single joint's peak current is in the 20-40A range, a single capacitor's capacitance should be between 68-100μF; when the peak current exceeds 40A, multiple capacitors need to be connected in parallel to achieve a total capacitance of 120μF or more, while sharing the ripple current to reduce ESR (equivalent series resistance) heating of individual capacitors.

· Second-stage high-frequency decoupling capacitor group: Arranged at the three-phase inverter bridge power input of the joint driver, adjacent to the collector/drain of the power switch. It is mainly used to provide high-frequency transient current, offset the parasitic inductance of the power loop, and control the power loop area to an extremely small range—the industry goal is to control the high-frequency power loop area within 1.5cm². This stage of capacitors requires extremely low ESL (equivalent series inductance) and ESR, so multiple small-capacitance (hundreds of pF range), low-ESR ceramic or film capacitors are typically connected in parallel to further reduce loop impedance at high frequencies, ensuring that when power devices need to rapidly switch large currents, the capacitors can instantly provide sufficient charge.

· π-type EMI filter unit: Arranged between the first-stage energy storage capacitor and the second-stage high-frequency decoupling capacitor, composed of a common-mode choke and X and Y capacitors on both sides, forming a π-type filter structure. The core function of this design is bidirectional EMI suppression—it suppresses EMI noise generated by the upstream power supply from entering the downstream joint driver, and also suppresses high-frequency noise generated during downstream power conversion from being conducted back through the power line to the upstream power supply, preventing noise from propagating throughout the entire power supply system.

3 Power conversion circuit architecture: high-performance three-phase inverter bridge + immunity design

The power conversion circuit is the core execution unit of the joint drive system. Its function is to efficiently and precisely convert the electrical energy of the DC bus into AC power to drive the motor. To cope with transient load impacts, this circuit requires targeted optimization in topology, switching device selection, gate drive logic, and snubber protection design.

（1） Topology

The joint motors of humanoid robots typically use three-phase brushless DC motors (BLDC) or permanent magnet synchronous motors (PMSM). The mainstream topology for the drive power stage of these motors is the three-phase voltage source inverter (VSI)—each phase leg consists of a high-side switch and a low-side switch in series, corresponding to the three-phase armature windings of the motor. By controlling the conduction and turn-off sequence and timing of the three-phase bridge switches, a rotating magnetic field with a 120° phase difference between phases can be generated in the motor's three-phase windings, driving the rotor to rotate.

To cope with high-frequency, high-current transient load conditions, the engineering implementation of this topology requires special attention to two points: first, the layout and routing of power devices must adopt a compact and symmetrical "straight-line" or "T-shaped" layout, minimizing the length of high-current traces and controlling the parasitic inductance of the power loop to an extremely low level; second, the upper and lower bridge switches of the inverter bridge must be set adequate deadtime to prevent shootthrough, which would create a direct bus shortcircuit—this dead time needs to be precisely configured by weighing switching losses and EMI noise.

（2） Switch device selection

The switching device is the core component of the three-phase inverter bridge. Its performance directly determines the dynamic response capability, efficiency, and reliability of the joint drive system. The switching speed and conduction loss performance limits of silicon-based MOSFETs can no longer meet the high-frequency, high-current transient load conditions under a 48V bus—therefore, the mainstream industry solution uses gallium nitride (GaN) high electron mobility transistors (HEMTs) as the core switching devices of the three-phase inverter bridge.

GaN devices' characteristics precisely match the core requirements of humanoid robot joint drives:

· High power density: The specific resistance (the product of chip area and on-resistance) of GaN devices is far lower than that of silicon-based MOSFETs with the same voltage rating—this means that under the same on-resistance requirement, GaN devices can be made with smaller chip areas. Combined with integrated half-bridge package designs, all power switching devices of the three-phase inverter bridge can be integrated into an extremely small volume, perfectly adapting to the narrow installation space constraints inside humanoid robot joints.

· Low switching loss: The reverse recovery charge (Qrr) of GaN devices is nearly zero, meaning that under high-frequency switching conditions, the reverse recovery loss of the device can be greatly reduced—compared with silicon-based MOSFETs of the same voltage rating, the switching loss of GaN devices can be reduced by approximately 70%. This characteristic not only improves the overall system energy efficiency but also simplifies the thermal design of the joint drive unit, further reducing the volume of the joint module.

· High voltage withstand capability: The voltage ratings of mainstream GaN devices now cover the 100V and above range—fully adapting to the transient spike requirements of a 48V bus voltage. Even if the bus voltage briefly rises due to regenerative impacts, the power devices will not be easily broken down.

In terms of specific device selection, the rotary actuator drive circuit of a certain brand's Ox model uses 100V-rated GaN power devices, paired with an integrated gate driver solution. Reference designs from semiconductor manufacturers also confirm this selection logic—for example, a certain MOSFET manufacturer's reference design uses three GaN half-bridge drivers, each with a voltage rating of 100V and a continuous operating current capability of 35A. The peak current level it can withstand is sufficient to cover the transient requirements of joint motors on a 48V bus.

（3） Gate drive circuit optimization

GaN devices have extremely fast switching speeds, which is their advantage, but it also brings new design challenges—if the gate drive circuit design is not sufficiently refined, the high-frequency noise generated during switching can cause oscillation in the device's gate voltage, and even trigger the Miller effect leading to false turn-on of the bridge arm, ultimately reducing system reliability. Therefore, the design of the gate drive circuit is a key aspect of GaN device application, and a precise balance must be achieved between switching speed and EMI noise.

The optimization details of the solution include:

► Use a dedicated isolated gate driver: This type of driver can provide sufficiently large gate drive current to quickly charge and discharge the GaN device's gate capacitance, fully leveraging GaN's high switching speed advantage. At the same time, the isolation design electrically isolates the drive control signal side from the high-voltage power side, suppressing noise transmission from the power side to the control side and protecting the downstream control circuitry. In terms of selection, the driver's peak source/sink current capability must be no less than 4A, and it must have transmission delay parameters matched to the GaN device's switching characteristics—many integrated GaN half-bridge drivers integrate the gate driver, GaN switches, bootstrap supply diode, and related logic protection circuits into a single package, which not only reduces the parasitic inductance of the drive loop but also greatly simplifies circuit design.

► Precisely configure gate resistors: In the gate drive loop, a low-value resistor must be connected in series to fine-tune the GaN device's switching speed—the smaller the resistance, the faster the switching speed and the lower the switching loss, but the greater the high-frequency EMI noise intensity; if the resistance is too large, although EMI noise is significantly reduced, the switching speed decreases and switching loss increases substantially. In practical engineering, the turn-on resistor and turn-off resistor are typically set to different precise values—for example, the turn-on resistor set to 5Ω and the turn-off resistor set to 2.2Ω, balancing switching loss and EMI noise levels through the difference in their resistance values.

► Minimize the parasitic inductance of the drive loop to the greatest extent: The routing of the drive loop must be as short as possible and placed as close as possible to the gate and source pins of the power switch. In PCB layout, the power switch, gate driver, decoupling capacitors, and related components need to be concentrated near the power loop to reduce the loop area. Some solutions also place high-frequency decoupling capacitors close to the gate driver's power pins to further reduce the parasitic inductance of the drive loop.

► Add Miller effect suppression circuitry: To prevent the risk of the GaN device's gate voltage being unexpectedly raised due to noise coupled through the parasitic capacitance between gate and drain during switching, which could cause false turn-on of the bridge arm, a Zener diode (Leiditech BZT52C5V6S, 5.6V, 200mW, <200pF) and a low-resistance Schottky diode (Leiditech LBAT40HT1G, 0.37V, 200mW, <40pF) are typically connected in parallel between the gate and source to suppress the effects of Miller-induced current.

（4） Snubber and protection circuit design

► RCD snubber circuit: On each switch of the three-phase inverter bridge, an RCD snubber circuit composed of a resistor, capacitor, and diode must be connected in parallel. Its function is to provide an energy release path for the parasitic inductance of the power loop during the switch turn-off process, slowing the voltage rise rate across the switch and reducing the amplitude of the turn-off spike voltage. The capacitor absorbs the spike energy, the resistor limits the capacitor's discharge current, and the diode (Leiditech DSK35, 50V, 3A) provides a freewheeling path for the inductor energy. In terms of parameter selection, it must be precisely matched according to the switch's performance and the parasitic inductance of the power loop—typically the resistor value is around 10Ω and the capacitor value is around 1nF.

► TVS overvoltage clamping: Between the positive and negative terminals of the DC bus, a high-power TVS (Leiditech 5.0SMDJ58CA, Vrwm 58V, 5000W) must be connected in parallel—its response speed reaches the picosecond level, allowing it to absorb spike energy from the bus in an extremely short time and clamp the bus voltage within the withstand voltage threshold of the power switch, preventing damage to the power switch from transient overvoltage.

► Current detection and fast protection circuit: At the output of the three-phase inverter bridge, high-precision, high-bandwidth phase current detection elements must be installed—typically low-cost shunt resistors or non-invasive Hall current sensors, combined with high-precision current sampling amplifiers, converting the three-phase current signals in the motor windings into voltage signals for transmission to the downstream control circuit. The control circuit continuously compares the detected current with a preset overcurrent protection threshold. Once an overcurrent fault is detected, it immediately triggers PWM signal blocking within a few microseconds, turning off the switches of the three-phase inverter bridge and limiting the impact of the fault to the smallest possible scope.

四.System control architecture: multi-dimensional coordinated suppression of transient loads

Hardware circuits are the fundamental guarantee for coping with transient load impacts, but without supporting control algorithms for real-time dynamic adjustment, the performance limits of the hardware cannot be fully utilized. Therefore, mainstream solutions adopt a comprehensive software control logic of "multi-axis coordinated energy management + motor drive current loop dynamic control + regenerative energy recovery," working in conjunction with the hardware circuits to collectively keep the impact of transient loads within the system's tolerable range.

1. Multi-axis coordinated energy management control

In multi-joint coordinated operation, the system-level main control unit or power management controller collects the real-time operating status of all joint drivers through a high-speed real-time communication bus—including key parameters such as DC bus voltage, motor phase current, motor speed, output torque, and joint position and angular velocity. It identifies in real time the number and power demand of joints currently in motoring state (absorbing power) and generating state (feeding back regenerative power), dynamically coordinates the power flow on the bus, and reasonably allocates the system's transient power demand among the hybrid energy storage unit and each joint.

The core of this logic is the "energy router" scheduling mechanism: priority is given to directly supplying the regenerative energy generated by joints in generating state to joints in motoring state through the bus; if the amount of regenerative energy exceeds the demand of the motoring joints, the excess energy is allocated to the supercapacitor for storage; if the supercapacitor is already fully charged and there is still excess regenerative energy, the braking unit is activated to dissipate the excess energy as heat. Conversely, when there is a large power gap on the bus, the supercapacitor is commanded to release stored energy through the bidirectional DC/DC converter to supplement the bus power gap, preventing large bus voltage drops.

This solution requires a high-real-time communication bus as support. Therefore, joint drivers and the main control unit typically communicate via industrial Ethernet buses such as EtherCAT or CANopen, with command transmission cycles controlled within the millisecond level to ensure the precision of coordinated power control among joints.

2 .Motor drive control algorithm optimization

The motor drive control algorithm needs to have extremely fast dynamic response capability—at the instant of load current mutation, it must quickly adjust the output current to offset the impact of load fluctuations. The solution adopts a dual closed-loop control architecture based on field-oriented control (FOC), namely cascaded control of position loop, speed loop, and current loop, optimizing the dynamic response of the current loop to smooth out current fluctuations caused by transient loads.

The technical details of this solution are: the FOC algorithm decomposes the motor's stator current vector into the excitation current component (Id) that generates the magnetic field and the torque current component (Iq) that generates torque. By independently and precisely controlling the magnitudes of these two components, precise adjustment of the motor's output torque is achieved. The optimization points targeting transient loads are mainly three:

► Dynamic current limiting: Based on the rotor's real-time position, motor temperature, and bus voltage fluctuation amplitude, dynamically adjust the given thresholds for d/q axis currents, limiting the motor's phase current peak within the safe operating area of the power switches, avoiding overcurrent that exceeds the power switches' carrying capacity.

► Load observer and feedforward compensation: A load torque observer is introduced into the algorithm. By collecting data such as motor phase current, rotor position, and speed, it estimates the joint's real-time load torque. This estimated value is used as a feedforward quantity, superimposed onto the current loop's given value, to compensate for load changes in advance—this can significantly reduce the motor speed drop amplitude under transient loads, enhancing the system's dynamic response capability.

► Real-time dead time optimization: Based on the motor's current magnitude and direction and the switches' actual switching characteristics, the dead time of the upper and lower bridge arms of the three-phase inverter bridge is adjusted in real time. On the premise of avoiding bridge shoot-through, the dead time duration is shortened as much as possible—this can reduce switching losses to a certain extent while also reducing current waveform distortion caused by dead time and reducing torque fluctuations caused by harmonics.

3. Regenerative energy recovery and dissipation management

As mentioned earlier, when a joint motor is in a braking state or is driven in reverse by a load, it converts mechanical energy into electrical energy and feeds it back to the DC bus. If this regenerative energy is not properly handled, it will directly cause an abnormal rise in bus voltage, leading to overvoltage damage to power devices. Therefore, a comprehensive regenerative energy management mechanism must be established to achieve in-situ balancing or safe dissipation of regenerative energy.

Under the common DC bus architecture, the priority of this control logic is: first, distribute the regenerative energy directly to other accelerating joint motors through the bus to achieve in-situ energy balancing; if the amount of regenerative energy exceeds the demand of other joints, then charge the excess energy into the supercapacitor module through the bidirectional DC/DC converter; if the supercapacitor is already fully charged and there is still excess regenerative energy, activate the braking unit—connect a braking resistor to the bus to dissipate the excess energy as heat, preventing the bus voltage from continuing to rise. The control logic of the braking unit adopts hysteresis control: when the bus voltage exceeds the set upper threshold, the braking resistor is connected to the bus; when the voltage falls below the set lower threshold, the braking resistor is disconnected to avoid unnecessary energy loss.

Schematic diagram of the humanoid robot DC power supply system circuit

(To simplify the schematic, only the circuit diagrams of 2 Knee Actuators and 2 Elbow Actuators are shown; the circuit diagrams of other Actuators are similar.)

四.Leiditech protection solution component selection list

五. Conclusion