In this paper we describe a Bayesian approach to model selection and state estimation for sensor-based robot tasks. The approach is illustrated with a hybrid model-state estimation example from forcecontrolled autonomous compliant motion: simultaneous (discrete) contact formation recognition and estimation of (continuous) geometrical parameters. Previous research in this area mostly tries to solve one of the two subproblems, or treats the contact formation recognition problem separately, avoiding integration between the solutions to the contact formation recognition and the geometrical parameter estimation problems. A more powerful hybrid model, explicitly modeling contact formation transitions, is developed to deal with larger uncertainties. This paper demonstrates that Kalman filter variants have limits: iterated extended Kalman filters can only handle small uncertainties on the geometrical parameters, while the non-minimal state Kalman filter cannot deal with model selection. Particle filters can handle the increased level of model complexity. Explicit measurement equations for the particle filter are derived from the implicit kinematic and energetic constraints. The experiments prove that the particle filter approach successfully estimates the hybrid joint posterior density of the discrete contact formation variable and the 12-dimensional, continuous geometrical parameter vector during the execution of an assembly task. The problem shows similarities with the well-known problems of data association in simultaneous localization and map-building (SLAM) and model selection in global localization.KEY WORDS-Bayesian model selection, state estimation, sensor-based robot tasks, autonomous compliant motion, simultaneous contact formation recognition and estimation of geometrical parameters, data association, SLAM, particle filter, hybrid joint density 1. In this paper we use the term "parameter" in the sense of a static state.

This paper presents a contribution to programming by human demonstration, in the context of compliant-motion task specification for sensor-controlled robot systems that physically interact with the environment. One wants to learn about the geometric parameters of the task and segment the total motion executed by the human into subtasks for the robot, that can each be executed with simple compliant-motion task specifications. The motion of the human demonstration tool is sensed with a 3-D camera, and the interaction with the environment is sensed with a force sensor in the human demonstration tool. Both measurements are uncertain, and do not give direct information about the geometric parameters of the contacting surfaces, or about the contact formations (CFs) encountered during the human demonstration. The paper uses a Bayesian sequential Monte Carlo method (also known as a particle filter) to do the simultaneous estimation of the CF (discrete information) and the geometric parameters (continuous information). The simultaneous CF segmentation and the geometric parameter estimation are helped by the availability of a contact state graph of all possible CFs. The presented approach applies to all compliant-motion tasks involving polyhedral objects with a known geometry, where the uncertain geometric parameters are the poses of the objects. This work improves the state of the art by scaling the contact estimation to all possible contacts, by presenting a prediction step based on the topological information of a contact state graph, and by presenting efficient algorithms that allow the estimation to operate in real time. In real-world experiments, it is shown that the approach is able to discriminate in real time between some 250 different CFs in the graph.

Abstract. This work presents a comparison of decision making criteria and optimization methods for active sensing in robotics. Active sensing incorporates the following aspects: (i) where to position sensors, and (ii) how to make decisions for next actions, in order to maximize information gain and minimize costs. We concentrate on the second aspect: "Where should the robot move at the next time step?". Pros and cons of the most often used statistical decision making strategies are discussed. Simulation results from a new multisine approach for active sensing of a nonholonomic mobile robot are given.

No abstract

No abstract