Model-based Optimization for Robotics

Abstract

Nature has demonstrated an incredible ability to optimize the designs and behaviors of humans, animals, and plants. Many areas of engineering have tried to copy the optimality principles of nature or optimize technical properties of systems, and mathematical optimization has been a very helpful tool in this context. To promote the use of optimization methods in robotics, the IEEE Robotics and Automation Society (RAS) Technical Committee (TC) on Model-Based Optimization for Robotics was founded and approved by the IEEE RAS Technical Activities Board and the RAS Administrative Committee in October 2012 with Cochairs Katja Mombaur, Abderrahmane Kheddar, Kensuke Harada, Thomas Buschmann, and Chris Atkeson. The TC focuses on optimization in robotics in the very general sense and, in particular, on the development and application of optimization methods for the design of robots and the generation and control of dynamic behaviors in robotics as well as their practical implementation. Optimization approaches are, in principle, applicable to any type of robot or robot behavior but are most interesting for robots with complex structures and dynamics, e.g., humanoid robots, or for high speeds, e.g., in mobile or agile manipulation (Figures 1–3). Optimization problems in robotics range from different tasks related to model identification (such as parameter estimation, optimum experimental design, inverse optimal control, and so on) over design optimization to optimal behavior generation (offline optimal control, model predictive control/online optimal control). Recent developments in optimization algorithms and, in particular, real-time optimization make a wider application of algorithmic optimization a realistic option even for real-time control in complex robotic applications in the near future. There is a growing interest in optimization in the robotics community for behavior generation and control, as recent publications and workshops show. However, it can be observed that the level of optimization techniques used is often far behind the current state of the art within the optimization community and that only very simple optimization problems are solved. This prevents the optimization potential of existing robot platforms from being fully exploited, and current designs are often suboptimal. But some papers show very interesting optimization solutions for the most advanced robotic systems, even for complex tasks with impacts or varying constraints, such as weight lifting or fast walking. At the same time, it is still a major challenge to actually apply these optimal solutions robustly to real robots and to handle modeling errors and uncertainties. Important research areas in the scope of this TC include: ● the optimization-based generation of robot trajectories using dynamical models of the robot and its environment ● improving the behavior style of robots by optimization, in particular for humanoid robots (induce natural behavior) ● online motion control using realtime model-based optimization and model predictive control/receding horizon control ● optimizing the design of robots for given tasks (parameter optimization and structural optimization) ● the development of appropriate dynamical models for offline and online optimization ● learning/improving models during optimization ● inverse optimal control techniques for the identification of objective functions ● robust optimal control and the refinement of optimal control policies based on actual experience ● the combination of optimization and machine-learning approaches ● the combination of optimization and path-planning methods.