Talk title: Safe OnGO-VIC: Online Gain Optimization for Variable Impedance Control with Control Barrier Functions
Abstract: Variable impedance control plays an important role in contact-rich manipulation tasks. It enables the robot to interact with unknown environments safely. Besides, it provides the flexibility to adapt the impedance according to different task objectives and scenarios. While many studies focus on designing the variable impedance control policies by adaptive control and reinforcement learning methods, these approaches are typically over specific to one task and not time-efficient to obtain. To deal with those problems, we present an efficient Safe Online Gain Optimization method for Variable Impedance Control (Safe OnGO-VIC). By reformulating the dynamics of impedance control as a control-affine system, in which the impedance gains are the inputs, we provide a novel perspective to understand the variable impedance control and innovatively formulate an optimization problem for online gain adaptation.
Moreover, the control barrier function (CBF) is incorporated into our gain optimization framework for safety guarantee. The proposed algorithm was experimentally validated on three manipulation tasks: 1) collision avoidance, 2) board contact and 3) surface wiping. The comparison results with a constant gain baseline and an adaptive control law prove that the proposed algorithm can achieve better performance efficiently, and it is more generalizable for different scenarios.