Actuator Design and Edge AI Improved Motion Quality for Humanoid Robots

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Broadband motor control, GaN power stages, and edge AI improve the motion quality of humanoid robots—despite limited installation space and within the scope of growing safety requirements.

Unlike industries with established architectures and interfaces, the design of a humanoid robot (Figure 1) requires different approaches depending on the application’s requirements, dimensions, and performance specifications. Given the variety of design concepts, OEMs and system integrators must explore various strategies to reduce space requirements while improving efficiency and performance. This, in turn, can significantly increase design complexity and accordingly extend development time.

This article discusses how advancements in broadband motor control, gallium nitride power stages, and edge AI enhance actuator capabilities by providing flexibility without relying on industry-wide standards. Space constraints and functional safety aspects are two additional topics briefly addressed.

Broader Motor Control

The movements of humanoid robots can quickly become unstable, inefficient, and unsafe if position, speed, and torque are not controlled precisely and responsively. Broadband motor controls aid in achieving smooth, human-like movements by enabling the algorithms executed by motor controls and embedded processors to continuously adapt the behavior of the actuators to changing loads and external forces.

For the implementation of broadband motor control, microcontrollers (MCUs) are required, such as the MCUs from the C2000 series, which are designed for real-time control. These components support the fast processing of current, speed, and position control loops with low, deterministic latency.

To optimize the performance of the humanoid motion system, large control bandwidths are required, such as:

Current control bandwidth over 5 kHz for responsive torque control;

Speed control bandwidth over 1 kHz in the interest of smooth motion profiles; or

Position control bandwidth over 1 kHz for precise joint positioning.

Achieving these bandwidths also requires the network to transmit position and speed setpoints from the MCU with minimal, deterministic latency. EtherCAT supports motion update rates between 4 and 32 kHz and a transmission bandwidth of 100 MBit/s, allowing the setpoint transmission to keep up with the bandwidths of the inner control loops required by real-time MCUs—and this for all joints simultaneously and without performance drops when additional joints are added.

Efficiency and Thermal Properties of the Overall System

When the control bandwidth is increased across multiple joints, power loss and cooling requirements grow, making efficiency a critical design criterion. After all, humanoid robots can contain dozens of simultaneously active actuators, so even minor efficiency deficiencies in individual joints can quickly add up, leading to excessive heat generation, which limits operating time and reduces performance.

If drive efficiency can be improved through advanced power technologies like GaN, this benefits efficiency as well as thermal properties. GaN power stages can achieve efficiencies of over 99%, compared to only 95% in silicon-based designs. This represents a significant improvement in power density and thermal management. Using GaN, developers can implement switching frequencies of over 100 kHz, enabling the use of smaller passive components while keeping EMI levels low.

When developing systems for humanoid robots, designers should aim for a power density of more than 5 W/cm³ at the actuator level, and the continuous power should be scaled according to the requirements of each joint. While finger joints are likely to range between 5 and 20 W, shoulder and elbow joints typically require between 100 and 750 W of continuous power. GaN-based half-bridge power stages with integrated drivers assist in achieving these high power densities.

Thermal behavior can also be influenced by system integration. Among other things, the integration of control, sensors, and power management makes it possible to increase overall efficiency at the joint level, which is becoming increasingly important given the growing complexity of humanoid robots.

Improvements through Edge AI

Edge AI plays an important, though supportive, role at the actuator level by monitoring and protecting operations at the joint level. By analyzing current, torque, and load, AI can detect early signs of wear and abnormal behavior. This predictive maintenance identifies potential failures before they can adversely affect system operation.

The use of Edge AI in actuators requires streamlined neural network architectures to meet the constraints faced by real-time MCUs in terms of computing power, memory capacity, and latency. In motion systems, feedback signals such as phase current, bus voltage, position, and speed are evaluated by an Edge AI model to detect system-level issues, enabling the identification of early signs of motor failures or wear in bearings or gear reducers. The data collected during system monitoring can also be utilized via machine learning to fine-tune control and diagnostic parameters. This helps maintain dynamic performance and consistency regardless of wear, drift, and changing operating conditions, which becomes increasingly important as components age.

Edge AI also supports local protective functions. By estimating torque and load conditions in real time, a robot's actuators can detect when they are approaching operational limits. They can then take corrective actions, such as limiting output power to prevent damage. These capabilities enhance the safety of interactions between humanoid robots and humans.

Implementation of a Compact Actuator Design

The limited space presents another fundamental challenge, as the joints in humanoid robots must deliver high torque and respond quickly despite small dimensions.

To maintain human-like dimensions, a high power density is necessary. The ideal configuration for the elbow joint of a humanoid robot, for example, requires a diameter between 50 and 100 mm.

This high level of compactness is not achievable without integration. Actuator designs must combine logic, sensing, and communication in a confined space while still providing acceptable thermal properties. Therefore, design decisions vary depending on the function and size of the respective joint and the available space. Figure 2 exemplifies a compact test board for the 48V/1kW motor control system of a robot joint with a 70 mm diameter.

Compliance with functional safety requirements

Unlike cobots and industrial robots, there are currently no established standards specifically applicable to humanoid robots. However, standardization organizations are very likely to develop safety specifications for humanoid robots in the future as market demand grows and the number of viable use cases increases.

Until safety requirements are established, developers of humanoid robots should thoroughly examine their current design concepts to reduce future modification costs. Standards such as ISO13849 or IEC 61508 provide an outlook on potential future requirements.

Conclusion: Combine Integration and Safety in Thinking

Effectively addressing the challenges outlined above requires a systematic approach early in the design phase. This should include rigorous testing and validation protocols and be based on safety-certified MCU platforms, such as the AM2612 and F28P650 MCUs from Texas Instruments, to meet safety standards for interacting with humanoid robots.

* Han Zhang is an Applications Manager in the team for application-specific MCU solutions for industrial motor control at TI, focusing on motor control solutions for household appliances, robotics, and industrial automation applications.

Robotics

Humanoids Put to the Test