Automatic LQR tuning based on Gaussian process global optimization

Abstract

Abstract-This paper proposes an automatic controller tuning framework based on linear optimal control combined with Bayesian optimization. With this framework, an initial set of controller gains is automatically improved according to a pre-defined performance objective evaluated from experimental data. The underlying Bayesian optimization algorithm is Entropy Search, which represents the latent objective as a Gaussian process and constructs an explicit belief over the location of the objective minimum. This is used to maximize the information gain from each experimental evaluation. Thus, this framework shall yield improved controllers with fewer evaluations compared to alternative approaches. A sevendegree-of-freedom robot arm balancing an inverted pole is used as the experimental demonstrator. Results of two-and fourdimensional tuning problems highlight the method's potential for automatic controller tuning on robotic platforms.

Combining Simulations and Real-Robot Experiments for Bayesian Optimization of Bipedal Gait Stabilization

“…Deisenroth et al [9], for example, developed a Bayesian approach to tune a cart-pole system, whereas Berkenkamp et al [6] proposed to use Bayesian optimization to safely tune robotic controllers for quadrotors. Moreover, Marco et al [16] combined Bayesian optimization with optimal control to tune LQR regulators. In contrast to these approaches, Akrour et al [1] suggested to direct the optimization process by using a search distribution, however, the optimization loses expressibility on a global scope since it only optimizes locally.…”

Gait Learning for Soft Microrobots Controlled by Light Fields

“…BO with GPs has successfully been used, for example, for gait learning with bipedal walkers [9], quadrupedal robots [10], and in cm-scale hexapodal robots [11]. It has also been proposed for automatic feedback controller tuning [12]- [14]. A major strength of GPs is that they allow one to include existing information about the system in the form of a probabilistic prior.…”

“…We formulate the problem of gait learning in soft microrobots as a parametric controller tuning problem. This work builds on the automatic controller tuning methods proposed in [12], where an unknown controller cost function was learned in a data-efficient way using BO with GPs.…”

Data-Efficient Autotuning With Bayesian Optimization: An Industrial Control Study

“…BO for controller learning has recently also been suggested in [12], [20], [21], which include successful demonstrations in laboratory experiments. A discrete event controller is optimized for a walking robot in [12], and state-feedback controllers are tuned in [20] for a quadrotor and in [21] for a humanoid robot balancing a pole. Herein, we present results of applying BO for a typical control problem in the automotive industry (throttle valve control) and consider two types of control objectives, different from those in [12], [20], [21].…”

“…A discrete event controller is optimized for a walking robot in [12], and state-feedback controllers are tuned in [20] for a quadrotor and in [21] for a humanoid robot balancing a pole. Herein, we present results of applying BO for a typical control problem in the automotive industry (throttle valve control) and consider two types of control objectives, different from those in [12], [20], [21]. The proposed controller learning framework, which combines BO with ADRC, is different from the controllers in the mentioned references.…”