足式跑動模型之模型結構修正與軌跡優化

Model Structure Modification and Trajectory Optimization of Legged Running Templates

楊文杉(Wen-Shan Yang)

指導教授 ： 林沛群

國立臺灣大學/工學院/機械工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201900128

摘要

隨著科技的進步，機器人發展得相當迅速，也因為著電腦運算能力的突破，在近幾年來人工智慧的技術也大量的被導入機器人的領域之中，許多以往被視為困難甚至是無解的問題，都有很大的改善與突破，像是對於複雜系統往往難以建立模型。本論文從結合基礎模型與高斯過程迴歸為出發，在模擬上的探討以點接觸之SLIP模型與實驗室建立之具有滾動特性之R-SLIP模型為基礎模型，而真實系統力與壓縮量的關係不會完全的線性以及系統具有阻尼，透過收集系統狀態與該狀態下響應的差異，透過高斯過程迴歸來找出系統狀態與所需補償之響應的關係，讓經過訓練後的複合模型對於實際系統的動態有更好的預測性，並透過實際機器人的動態實驗，修正簡化模型以達到更好的動態預測。在模型改善後另一個難題則是軌跡的優化，由於實際機器人的硬體限制，因此軌跡都是以五次多項式去擬合模擬中穩定點的軌跡，而在模擬上五次多項式所擬合的軌跡並不會讓模型的動態收斂到原本的穩定點甚至無法收斂，因此這裡開發一套以五次多項式為基礎，透過最佳化的方式來決定出新的五次多項式係數。

關鍵字

動態模型 ； 高斯過程迴歸 ； 穩定性分析 ； 軌跡最佳化

並列摘要

With the advancement of technology, robotics has developed rapidly. Also credited to the breakthrough of computing power, artificial intelligence has been introduced into robotic fields. Many problems regarded as difficult or unsolvable previously, for example, building models for complex systems, have received great improvement and breakthrough. This research starts from combining basic models and Gaussian process regression, investigating point-contact SLIP model and rolling-contact R-SLIP model. However, in real systems, the relation between force and compression is not linear and the system is damped. By collecting the difference between the response of real system and the response of the model under the same condition, the compensation for the model to simulate the real system can be found by Gaussian Process Regression. The trained hybrid model is more capable of predicting the real dynamic system. On the other hand, by the experiment on the real robot, the more precise data of the real system is collected in order to achieve better dynamic predictability. Another issue of the model improvement is the optimization of trajectories. Owing to the constraint of the hardware on the real robot, the trajectories transmitted to the robot in the form of 5-th order polynomial fitting the trajectories of stable points. Nonetheless, in simulations, the fitted polynomial cannot make the model converge to the original stable point, or cannot converge sometimes. As a result, this research develops a 5-th order polynomial-based method deciding a new better set of coefficients of polynomial by optimization.

並列關鍵字

Dynamic Model ； Gaussian Process Regression ； Stability Analysis ； Trajectory Optimization

參考文獻

[1] R. Altendorfer, D. E. Koditschek, and P. Holmes, "Stability analysis of a clock-driven rigid-body SLIP model for RHex," The International Journal of Robotics Research, vol. 23, no. 10-11, pp. 1001-1012, 2004.

[2] M. M. Ankarali, N. J. Cowan, and U. Saranli, "TD-SLIP: A better predictive model for human running," Proceedings of Dynamic Walking, 2012.

[3] J. Rummel and A. Seyfarth, "Stable running with segmented legs," The International Journal of Robotics Research, vol. 27, no. 8, pp. 919-934, 2008.

[4] C.-J. Hu, C.-K. Huang, and P.-C. Lin, "A torque-actuated dissipative spring loaded inverted pendulum model with rolling contact and its use as the template for design and dynamic behavior generation on a hexapod robot," in Robotics and Automation (ICRA), 2015 IEEE International Conference on, 2015, pp. 5177-5183: IEEE.

[5] P. Holmes, R. J. Full, D. Koditschek, and J. Guckenheimer, "The dynamics of legged locomotion: Models, analyses, and challenges," Siam Review, vol. 48, no. 2, pp. 207-304, 2006.