High-Precision Trajectory Tracking in Changing Environments Through - PowerPoint PPT Presentation

High-Precision Trajectory Tracking in Changing Environments Through ℒ " Adaptive Feedback and Iterative Learning Karime Pereida, Rikky R. P. R. Duivenvoorden, and Angela P. Schoellig Additional contributions from Dave Kooijman ICRA Spotlight Talk May 30 th , 2017

Motivation Different Goal: achieve high systems tracking performance Varying payloads Wind Unknown and changing disturbances Varying topology Tracking error Varying weather 2 Karime Pereida

Objectives Goal: achieve high tracking performance Unknown and changing disturbances Tracking error System 1 System 2 Input Output Repeatable and Improve over reliable iterations High tracking behavior performance even if dynamics change 3 Karime Pereida

� Proposed Approach Repeatable and • Define a reference (desired) behavior. ℒ " Adaptive reliable Controller • Stay provably close to reference model. behavior : 𝑇𝑧𝑡𝑢𝑓𝑛 𝑃𝑣𝑢𝑞𝑣𝑢 − 𝑆𝑓𝑔𝑓𝑠𝑓𝑜𝑑𝑓 𝑁𝑝𝑒𝑓𝑚 𝑃𝑣𝑢𝑞𝑣𝑢 < 𝛿 ∝ ; • Zero tracking error not guaranteed. • Can compensate for systematic tracking Iterative Improve over errors. Learning iterations Controller • Learns through repetition. • Fast convergence. High tracking performance even • No re-learning if system or dynamics if dynamics change change. 4 Karime Pereida

Previous Work Iterative Learning Unknown and Controller changing Tracking error disturbances Input ℒ " Adaptive System 1 Output Controller Simulation results [6] B. Altın and K. Barton, “Robust iterative learning for high precision motion control through ℒ " adaptive feedback,” Mechatronics , vol. 24, no. 6, pp. 549–561, 2014. 5 Karime Pereida

Proposed Approach Reference model behavior Disturbances ℒ " Adaptive System 1 System 2 Input Output controller Improve tracking over iterations Updated Input Iterative Learning Controller 6 Karime Pereida

Results Reference model behavior Disturbances Proportional- ℒ " Adaptive Proportional- System 1 Input Output Integral- Derivative controller Derivative Improve tracking over iterations Updated Input Iterative Learning Controller 7 Karime Pereida

Results Unknown and changing disturbances: Wind Contributor: Dave Kooijman 8 Karime Pereida

Results Unknown and changing disturbances: Wind Contributor: Dave Kooijman 9 Karime Pereida

Results Transfer learning: no need to relearn Contributor: Dave Kooijman 10 Karime Pereida

Results Transfer learning: no need to relearn Contributor: Dave Kooijman 11 Karime Pereida

Summary Reference model behavior Repeatable and Disturbances reliable ℒ " Adaptive behavior System 1 Output Input controller Improve over iterations Improve tracking over iterations High tracking Updated Input Iterative Learning Controller performance even if dynamics change • Extension work K. Pereida, D. Kooijman, Rikky R. P. R. Duivenvoorden, and Angela P. Schoellig, “Transfer Learning for High-Accuracy Trajectory Tracking Through ℒ " Adaptive Feedback and Iterative Learning” , submitted to International Journal of Adaptive Control and Signal Processing. - Transfer learning from simulation to real system. - Use reference model to calculate input. 12 Karime Pereida

Thank you! Karime Pereida Pérez <karime.pereida@robotics.utias.utoronto.ca>