Bayesian optimization for robotics and control
We have developed several algorithms for Bayesian optimization, most of them applied to robotics, reinforcement learning and control problems.
Distributed Bayesian optimization
Javier Garcia-Barcos and Ruben Martinez-Cantin (2019) Fully Distributed Bayesian Optimization with Stochastic Policies. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, {IJCAI-19}., pages 2357-2363. (PDF) (BibTeX)
Spartan Bayesian optimization
Ruben Martinez-Cantin (2019) Funneled Bayesian Optimization for Design, Tuning and Control of Autonomous Systems. IEEE Transactions on Cybernetics, 49(4):1489-1500. (PDF) (BibTeX)
Ruben Martinez-Cantin (2017) Bayesian Optimization with Adaptive Kernels for Robot Control. In Proc. of the IEEE International Conference on Robotics and Automation., pages 3350-3356. (PDF) (BibTeX)
Robust Bayesian optimization
Ruben Martinez-Cantin, Kevin Tee and Michael McCourt (2018) Practical Bayesian optimization in the presence of outliers. In International Conference on Artificial Intelligence and Statistics (AISTATS), PMLR 84., pages 1722-1731. (PDF) (BibTeX)
Ruben Martinez-Cantin, Kevin Tee and Michael McCourt (2017) Policy Search using Robust Bayesian Optimization. In Neural Information Processing Systems (NIPS) Workshop on Acting and Interacting in the Real World: Challenges in Robot Learning.. (PDF) (BibTeX)
Unscented Bayesian optimization for grasping
Bayesian optimization is a great tool to find optimal grasps for its sample efficiency, requiring many trials. However, the robot might have trouble repeating exactly the same grasp in the same position. Thus, it is necessary to perform a sensitivity analysis of the solution.
Unscented Bayesian optimization allows to perform a simple analysis each iteration, driving the optimization towards safe and repeatable grasps.
José Nogueira, Ruben Martinez-Cantin, Alexandre Bernardino and Lorenzo Jamone (2016) Unscented Bayesian Optimization for Safe Robot Grasping. In Proc. of the IEEE/RSJ Int. Conf. on Intelligent Robots and Systems.. (PDF) (BibTeX)
Joao Castanheira, Pedro Vicente, Ruben Martinez-Cantin, Lorenzo Jamone and Alexandre Bernardino (2018) Finding safe 3D robot grasps through efficient haptic exploration with unscented Bayesian optimization and collision penalty. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2018).. (PDF) (BibTeX)