TL;DR
This paper introduces an Iterative Learning Control (ILC) approach to precisely track motion trajectories for a novel soft crawling robot powered by dielectric elastomer actuators (DEAs). By leveraging past performance, the ILC method effectively overcomes the modeling challenges inherent in soft robotics, enabling robust and accurate locomotion.
What Problem Does This Paper Address?
Soft robots, with their inherent flexibility and adaptability, hold immense promise for applications in unstructured and delicate environments. However, their complex electromechanical coupling and viscoelastic properties make accurate dynamic modeling a significant hurdle. This difficulty in modeling has historically limited the development of precise motion control strategies for soft robots, hindering their practical deployment. This paper specifically tackles the challenge of achieving accurate motion trajectory tracking for a circular soft crawling robot, which relies on DEAs for actuation and electroadhesion actuators for locomotion. The goal is to develop a control method that can achieve high-performance tracking despite the inherent uncertainties in the robot's dynamics.
Methods
The core of this paper is the application of Iterative Learning Control (ILC) to the soft crawling robot. ILC is a control strategy well-suited for repetitive tasks, where the system learns from its past performance to improve its control inputs over successive iterations. The authors propose a "D² type" ILC algorithm, which is a higher-order ILC law designed to handle systems with a relative degree greater than one, a characteristic of the DEA-driven robot.
To facilitate the ILC design and analysis, a knowledge-based modeling framework is employed. This involves creating a simplified dynamic model of the DEA, drawing parallels to biological muscle models using spring-dashpot elements. This approach aims to capture the essential dynamics without succumbing to the overwhelming complexity of a fully detailed model. The ILC controller itself is designed to be partially model-free, meaning it can achieve excellent tracking performance even with uncertainties in the identified model parameters. The learning gain ($\gamma$) is the primary parameter to be tuned, simplifying the controller setup.
Datasets
Not discussed in the paper. The paper focuses on system identification and experimental validation using a custom-built robot.
Key Insights and Ideas
- ILC as a Solution for Soft Robot Modeling Challenges: The paper strongly advocates for ILC as a powerful tool for controlling soft robots. Its ability to learn from past iterations and adapt to system uncertainties makes it a practical alternative to complex model-based control.
- Knowledge-Based Modeling for Simplification: The use of a knowledge-based model, inspired by biological muscle structures, offers a pragmatic way to represent the DEA's dynamics without getting bogged down in excessive detail. This simplification is crucial for controller design.
- Effective Trajectory Tracking: The results demonstrate that ILC can achieve excellent trajectory tracking, even for relatively complex paths, by iteratively refining the control inputs. This is a significant step towards enabling more sophisticated soft robot behaviors.
- Model-Free Advantage: The partially model-free nature of the ILC controller is highlighted as a key advantage, reducing the burden of precise system identification and allowing for adaptability to variations.
Key Results
- Simulation: Simulations showed that the ILC scheme effectively tracks both simple "S-type" and more complex "M-shaped" trajectories. After 50 iterations, the tracking error was significantly reduced, with the Maximum Absolute Error Rate (MAER) shrinking to less than 2% for the M-shaped trajectory.
- Experiment: Real-time experiments confirmed the effectiveness of the ILC. The tracking error, initially large in the first iteration, was gradually reduced and almost eliminated after 20 iterations, with the MAER falling within 10%. The entire training process for a robust control solution took approximately 10 minutes.
Limitations and Future Work
- Model Simplification: While the knowledge-based model simplifies the design, it is acknowledged as a simplified version of the actual DEA model. The paper notes that deviations between simulated and actual outputs exist, partly due to model uncertainties and the inherent viscoelasticity of DEAs.
- No Comparative Analysis: The paper does not include a comparative analysis with other control schemes, focusing solely on demonstrating the effectiveness of the proposed ILC.
- Future Work: The authors suggest that the developed soft robot, with its minimalist design, can be assembled with other components to realize more complex functional robots. The inherent isotropy of the circular structure also simplifies the analysis of motion characteristics, opening avenues for further research. The paper also implies that future work could explore more advanced ILC variants or hybrid control strategies to further enhance performance and robustness.
Summary updated