Method for controlling a robot and robot controller

1. A computer-implemented method for controlling a robot using control parameter values from a non-Euclidean control parameter space the method comprising:

performing a Bayesian optimization of an objective function representing a desired control objective of the robot over a control parameter space, wherein evaluation points of the objective function are determined by searching an optimum of an acquisition function in an iterative search, including, in each iteration:

updating a candidate evaluation point using a search direction in a tangent space of the parameter space at the candidate evaluation point,

mapping the updated candidate evaluation point from the tangent space to the parameter space, and

using the mapped updated candidate evaluation point as candidate evaluation point for a next iteration until a stop criterion is fulfilled;

controlling the robot in accordance with a control parameter value found in the Bayesian optimization;

determining the search direction according to the conjugate gradient method; and

determining a search direction for the mapped updated candidate evaluation point by modifying a gradient of the acquisition function at the mapped updated candidate evaluation point by a multiple of the search direction at the candidate evaluation point mapped to the tangent space of the parameter space at the mapped updated candidate evaluation point by parallel transport.

2. The method according to claim 1 , wherein the mapping of the updated candidate evaluation point from the tangent space to the parameter space is performed using an exponential map of the tangent space at the candidate evaluation point.

3. The method according to claim 1 , wherein the parameter space is a Riemannian manifold and wherein the Bayesian optimization uses a Gaussian process as surrogate model having a kernel dependent on an induced metric of the Riemannian manifold.

4. The method according to claim 3 , further comprising:

setting a parameter of the kernel to make the kernel positive definite.

5. The method according to claim 1 , wherein the updating of the candidate evaluation point includes a line search along the search direction.

6. The method according to claim 1 , wherein the parameter space is a sphere or a manifold of symmetric positive definite matrices.

7. The method according to claim 1 , wherein the objective function represents a desired position of a part of the robot.

8. The method according to claim 1 , wherein the parameter values represent stiffness or inertia or manipulability or orientation or pose.

9. A computer-implemented method for controlling a robot using control parameter values from a non-Euclidean control parameter space the method comprising:

performing a Bayesian optimization of an objective function representing a desired control objective of the robot over a control parameter space, wherein evaluation points of the objective function are determined by searching an optimum of an acquisition function in an iterative search, including, in each iteration:

updating a candidate evaluation point using a search direction in a tangent space of the parameter space at the candidate evaluation point,

mapping the updated candidate evaluation point from the tangent space to the parameter space, and

using the mapped updated candidate evaluation point as candidate

evaluation point for a next iteration until a stop criterion is fulfilled; and

controlling the robot in accordance with a control parameter value found in the Bayesian optimization;

wherein the parameter space is a Riemannian manifold and wherein the Bayesian optimization uses a Gaussian process as surrogate model having a kernel dependent on an induced metric of the Riemannian manifold;

calculating kernel matrices for multiple randomly sampled pairs of points of the parameter space and setting a parameter of the kernel to a minimum value for which the kernel matrices are positive definite.

10. The method according to claim 9 , further comprising:

determining the search direction according to the conjugate gradient method.

11. A robot controller configured to control a robot using control parameter values from a non-Euclidean control parameter space, the robot controller configured to:

perform a Bayesian optimization of an objective function representing a desired control objective of the robot over a control parameter space, wherein evaluation points of the objective function are determined by searching an optimum of an acquisition function in an iterative search, wherein the robot control is configured to, in each iteration:

update a candidate evaluation point using a search direction in a tangent space of the parameter space at the candidate evaluation point,

map the updated candidate evaluation point from the tangent space to the parameter space, and

use the mapped updated candidate evaluation point as candidate evaluation point for a next iteration until a stop criterion is fulfilled;

control the robot in accordance with a control parameter value found in the Bayesian optimization;

determine the search direction according to the conjugate gradient method;

determine a search direction for the mapped updated candidate evaluation point by modifying a gradient of the acquisition function at the mapped updated candidate evaluation point by a multiple of the search direction at the candidate evaluation point mapped to the tangent space of the parameter space at the mapped updated candidate evaluation point by parallel transport.

12. A non-transitory computer readable medium on which is stored instructions for controlling a robot using control parameter values from a non-Euclidean control parameter space, the instructions, when executed by a processor, causing the processor to perform the following steps:

performing a Bayesian optimization of an objective function representing a desired control objective of the robot over a control parameter space, wherein evaluation points of the objective function are determined by searching an optimum of an acquisition function in an iterative search, including, in each iteration:

updating a candidate evaluation point using a search direction in a tangent space of the parameter space at the candidate evaluation point,

mapping the updated candidate evaluation point from the tangent space to the parameter space, and

using the mapped updated candidate evaluation point as candidate evaluation point for a next iteration until a stop criterion is fulfilled;

controlling the robot in accordance with a control parameter value found in the Bayesian optimization;

determining the search direction according to the conjugate gradient method; and

determining a search direction for the mapped updated candidate evaluation point by modifying a gradient of the acquisition function at the mapped updated candidate evaluation point by a multiple of the search direction at the candidate evaluation point mapped to the tangent space of the parameter space at the mapped updated candidate evaluation point by parallel transport.