1. Automatic Snake Gait Generation Using ILQR
Snake robots have great potential for rescue and search because of their adaptability to cluttered and complex terrain. Currently snake locomotion is achieved through position control based on predefined motion patterns, i.e., snake gaits like serpenoid curves. Parameters used rarely transfer to a different environmental conditions and hand-tuning is always required. So we use iLQR to automatically generate snake gaits across different environment conditions, i.e., anisotropic Coulomb friction, viscous friction, drag force with added mass effect. (Here is link to the paper. A video that demonstrates the results is also attached.)
2. Thesis: From Model-Based to Data-Driven Discrete Iterative Learning Control
The concept of ILC began to flourish in 1984 motivated by robots doing repetitive tasks. ILC aims to achieve trajectory tracking accuracy on the reproducibility of the hardware. Using an a prior model, ILC improves the tracking accuracy iteratively using data measured in practice. Currently the study of ILC focuses on nonlinear systems and improvement of practical performance.
My research also focuses on these two aspects. For nonlinear systems, I proposed nonlinear ILC algorithms based on Carleman bilinearization and feedback linearization, the convergence rate of which are faster than ILC with linearized models. As for practical performance, I developed circulant filter to improve robustness and adaptive ILC to increase the learning rate.
I organized my studies from the perspective of data versus model, from model-based ILC to data-driven model-based ILC, and then data-driven model-free ILC. The model-free data-driven ILC is my first step to bridge reinforcement learning and ILC.
Here is the link to my thesis. Publications related to each chapter can be downloaded from this website.
3. Identification of Lung Mechanics for Ventilated Patients by Treating Respiratory Effort as an Unknown Periodic Disturbance
To diagnose/monitor some disease or measure the effects of treatments, we want to estimate the respiratory mechanics. However current methods are either invasive or need to interfere the ventilator maneuver.
So I investigated a system identification method for the estimation of lung mechanics with non-invasive and maneuver-free measurements. I used the system identification method that can cancel unknown periodic disturbances. The periodic disturbances are absorbed into a re-established multiple-step input-output model. This is my one-month internship project at Philips Research.
List of Publications
E. Hannigan, B. Song, G. Khandate, M. H. Heger, J. Yin, & M. Ciocarlie. “Automatic Snake Gait Generation Using Model Predictive Control”, arXivpreprint arXiv:1909.11204, 2019
B. Song, M. Phan, and R. W. Longman. “Feedback Linearization-based Discrete-Time Iterative Learning Control for Nonlinear Systems”, Advances in the Astronautical Sciences, 2019 (in press)
E. Hannigan, B. Song, G. Khandate, J. Yin, M. Haas Heger, and M. Ciocarlie. “SBP-Guided MPC to Overcome Local Minima in Trajectory Planning”, Workshop on Toward Online Optimal Control of Dynamic Robots, IEEE Intl. Conference on Robotics and Automation, Montreal, May 2019
G. Khandate, E. Hannigan, M. Hass Heger, B. Song, J. Yin, and M. Ciocarlie. “Algorithmic Gait Synthesis for a Snake Robot”, Workshop on Toward Online Optimal Control of Dynamic Robots, IEEE Intl. Conference on Robotics and Automation, Montreal, May 2019
B. Song, M. Phan, and R. W. Longman. “Data-Driven Model-Free Iterative Learning Control using Reinforcement Learning”, Advances in the Astronautical Sciences, 2018 (in press)
B. Song, M. Phan, and R. W. Longman. “Bilinearized Model-Based Discrete-Time Iterative Learning Control for Nonlinear Systems”, Advances in the Astronautical Sciences, 2018 (in press)
B. Song and R. W. Longman. “Modifying Iterative Learning Control to Increase Tracking Speed by Markov Parameter Updates”, Advances in the Astronautical Sciences, 158, 2307–2326, 2016
J. Shang, J. Yan, Z. Zhang, X. Huang, P. Maturavongsadit, B. Song, Y. Jia et al. “A hydrogel-based glucose affinity microsensor”, Sensors and Actuators B: Chemical, 237, 992-998, 2016
B. Song and R. W. Longman. “Circulant Zero-Phase Low-Pass Filter Design for Improved Robustification of Iterative Learning Control”, Advances in the Astronautical Sciences, 156, 2161–2180, 2015
X. Huang, C. Leduc, Y. Ravussin, S. Li, E. Davis, B. Song, D. Li et al. “A differential dielectric affinity glucose sensor”, Lab on a Chip 14, 2, 294-301, 2014
D. Li, H. Yu, X. Huang, B. Song, Y. Jia, Y. Ji, N. Li, J. Chen, K. Xu, and Q. Lin. “A microfluidic system with volume sensor and dielectric glucose sensor for continuous glucose monitoring”, The 17th International Conference on Solid-State Sensors, Actuators and Microsystems (TRANSDUCERS & EUROSENSORS XXVII), Barcelona, 365-368, 2013
S. Li, E. Davis, X. Huang, B. Song, R. Peltzman, D. Sims, Q. Lin and Q. Wang. “Synthesis and Development of Poly(N-hydroxyethyl acrylamide)-ran-PAAPBA (PHEAA-ran-PAAPBA) Polymer Fluid for Potential Application in Affinity Sensing of Glucose”, Journal of Diabetes Science and Technology 5: 1060-1067, 2011