Estimation and Control of Motor Core Temperature with Online Learning of Thermal Model Parameters: Application to Musculoskeletal Humanoids (2407.08055v1)

Abstract: The estimation and management of motor temperature are important for the continuous movements of robots. In this study, we propose an online learning method of thermal model parameters of motors for an accurate estimation of motor core temperature. Also, we propose a management method of motor core temperature using the updated model and anomaly detection method of motors. Finally, we apply this method to the muscles of the musculoskeletal humanoid and verify the ability of continuous movements.

• The paper introduces an online learning method that dynamically updates thermal model parameters to improve motor core temperature estimation.
• It demonstrates reduced estimation errors and effective real-time anomaly detection in motor operations on musculoskeletal humanoids.
• The thermal control strategy maintains safe motor temperatures under high loads, enhancing the reliability and longevity of humanoid robots.

Estimation and Control of Motor Core Temperature with Online Learning of Thermal Model Parameters: Application to Musculoskeletal Humanoids

The paper presents a comprehensive approach for the estimation and control of motor core temperature in musculoskeletal humanoids using online learning algorithms. The authors address a critical challenge in humanoid robotics: managing motor temperatures to ensure continuous and reliable operation of robots, especially those with stringent physical constraints. The paper offers promising methods for accurate thermal management through a combination of online learning, thermal modeling, and anomaly detection, along with practical implementation on musculoskeletal humanoids.

Methodology and Contributions

The work introduces a novel online learning method to dynamically update the thermal model parameters of motors. This approach ensures precise estimation of the motor core temperature, which is crucial for maintaining operational stability and preventing overheating. The methodology includes:

1. Basic Thermal Model: The paper employs a two-resistor thermal model, encompassing motor core temperature ($c_1$) and motor housing temperature ($c_2$). Key thermal resistances and heat capacities between these components and the ambient environment are modeled.
2. Proposed Thermal Model: The authors extend the basic model, introducing online learning to continuously update thermal parameters ($P_{\{1, 2, 3, 4, 5\}\}). This dynamic model adapts to changes in motor conditions over time, improving the accuracy of \(c_{1}$ estimation.
3. Thermal Estimator: A real-time estimation algorithm for $c_{1}$ is proposed, leveraging sensor data from the motor housing temperature ($c_{2}$) and accounting for the applied motor load ($f$).
4. Online Learning Mechanism: The paper details a procedure for accumulating data and updating the thermal model parameters using backpropagation through time. This ensures that the model remains accurate despite variations in motor performance and ambient conditions.
5. Anomaly Detection: Utilizing the learned thermal parameters, the paper introduces a method for detecting anomalies in motor operation. Significant deviations in $P_{\{1, 2, 3, 4\}$ from their baseline values indicate potential issues such as sensor failure or mechanical faults.
6. Thermal Control: An optimization-based control mechanism dynamically adjusts the motor's maximum output to maintain $c_{1}$ within a safe range, ensuring reliable and extended operation of the robot.
7. Application to Musculoskeletal Humanoids: The methods are implemented on Musashi, a musculoskeletal humanoid robot, demonstrating the practical viability of the proposed approach. The paper showcases effective control of motor temperatures during continuous and high-stress operations.

Experimental Results

The experimental analysis includes simulations and real-world tests on a single muscle actuator and an array of muscle actuators in a humanoid robot. Key findings include:

• Simulation Experiment: The proposed online learning method significantly reduced the parameter estimation errors, with convergence achieved in approximately 1200 seconds. Evaluation showed that both $c_{1}$ and $c_{2}$ estimations improved post-learning.
• Anomaly Detection: The system successfully identified anomalies within 400 to 600 seconds under various fault conditions. This demonstrates the robustness of using thermal parameter deviations for fault detection.
• Thermal Control Efficiency: The thermal control method effectively maintained $c_{1}$ at or below the maximum permissible temperature, even under high-tension loads, thus validating the control strategy.
• Actual Robot Application: Applying the methods to Musashi's arm muscles demonstrated real-time adaptability and effective thermal management, evidenced by stable $c_{1}$ values during prolonged operations.

Implications and Future Work

The research holds substantial implications for the development and deployment of humanoid robots. Accurate thermal management extends the operational life and reliability of robots, critical for applications in hazardous environments, healthcare, and human-robot interaction. The proposed methodologies not only improve performance but also enhance safety through timely anomaly detection.

Future development could focus on integrating the thermal management system directly into motion planning algorithms, allowing for more nuanced and context-aware adjustments during complex tasks like ambulation and dexterous manipulation. Additionally, optimizing computational efficiency to enable higher frequency updates without compromising real-time responsiveness would further enhance the system's application scope.

In conclusion, this paper presents a robust solution for motor temperature management in humanoid robots, leveraging online learning and optimization to ensure precise and adaptive control. The methods proposed and validated in this paper represent significant advancements in the intersection of robotics, thermal modeling, and machine learning.