Unscented Bayesian optimization for safe robot grasping

Abstract

We address the robot grasp optimization problem of unknown objects considering uncertainty in the input space. Grasping unknown objects can be achieved by using a trial and error exploration strategy. Bayesian optimization is a sample efficient optimization algorithm that is especially suitable for this setups as it actively reduces the number of trials for learning about the function to optimize. In fact, this active object exploration is the same strategy that infants do to learn optimal grasps. One problem that arises while learning grasping policies is that some configurations of grasp parameters may be very sensitive to error in the relative pose between the object and robot end-effector. We call these configurations unsafe because small errors during grasp execution may turn good grasps into bad grasps. Therefore, to reduce the risk of grasp failure, grasps should be planned in safe areas. We propose a new algorithm, Unscented Bayesian optimization that is able to perform sample efficient optimization while taking into consideration input noise to find safe optima. The contribution of Unscented Bayesian optimization is twofold as if provides a new decision process that drives exploration to safe regions and a new selection procedure that chooses the optimal in terms of its safety without extra analysis or computational cost. Both contributions are rooted on the strong theory behind the unscented transformation, a popular nonlinear approximation method. We show its advantages with respect to the classical Bayesian optimization both in synthetic problems and in realistic robot grasp simulations. The results highlights that our method achieves optimal and robust grasping policies after few trials while the selected grasps remain in safe regions.

Bayesian Multiobjective Optimisation With Mixed Analytical and Black-Box Functions: Application to Tissue Engineering

“…Note: larger, positive values for the metrics in Eqs. (19)- (21) indicate better performance than random search.…”

“…We treat these in silico models as black boxes for the purpose of optimisation when we can simulate reactor conditions but cannot efficiently derive gradient or Hessian information. Relevant black-box optimisation methods include genetic algorithms [16], trust-region methods [17], and Bayesian optimisation [18]- [21]. Black-box methods give flexibility with respect to the black-box contents.…”

Bayesian Optimisation for Safe Navigation Under Localisation Uncertainty

“…The first is standard BO, which does not consider any uncertainty in the location estimates. The second is unscented Bayesian optimisation (UBO) [15], which considers execution noise by means of the unscented transform [26], but assumes that the location estimate of the observation is accurate.…”

“…Recent work [15] presented a method to apply BO to problems where the execution of a query is uncertain, such as robotic grasping. The authors propose querying BO's surrogate model using a Gaussian distribution by applying the unscented transform [26].…”

Manipulation of Unknown Objects to Improve the Grasp Quality Using Tactile Information

“…When an exact object model is not available it can be approximated using geometric primitives [ 13 ] or learning methods can be applied to transfer a successful grasp of a known object to novel objects [ 14 ]. Uncertainty on the object shape has been modeled as constraints in the grasp planner [ 15 ], or as a noise managed using probabilistic techniques [ 16 , 17 ]. When the model of the object is completely unknown, a haptic exploration of the object surface can be performed prior to compute the grasp [ 18 ].…”