Abstract
Our recent work on consensus complementarity control (C3) proposes a model predictive control algorithm for hybrid systems that make and break contact with their environment. C3 is based on the alternating direction method of multipliers (ADMM), that is capable of high-speed reasoning over potential contact events. Via a consensus formulation, the approach enables parallelization of the contact scheduling problem and is able to run at real-time rates. In this abstract, we build upon this with software improvements that increase the run-time by approximately 2 times over our previous implementation (achieving 25 Hz rate for a problem with 19 states, 12 complementarity variables and 0.5 seconds prediction horizon). Additionally we integrated our software with Drake [4]. The software parses models in URDF format into a non-smooth linear complementarity system and generates a controller that solves the optimal control problem. In addition to our previous work, we also validated our approach on a 3D manipulation example.
Adam Wei
EECS Ph.D. Candidate
I am interested in robotics, manipulation, machine learning, and optimization.