We present early pilot-studies of a new international project, developing advanced robotics to handle nuclear waste. Despite enormous remote handling requirements, there has been remarkably little use of robots by the nuclear industry. The few robots deployed have been directly teleoperated in rudimentary ways, with no advanced control methods or autonomy. Most remote handling is still done by an aging workforce of highly skilled experts, using 1960s style mechanical Master-Slave devices. In contrast, this paper explores how novice human operators can rapidly learn to control modern robots to perform basic manipulation tasks; also how autonomous robotics techniques can be used for operator assistance, to increase throughput rates, decrease errors, and enhance safety. We compare humans directly teleoperating a robot arm, against humansupervised semi-autonomous control exploiting computer vision, visual servoing and autonomous grasping algorithms. We show how novice operators rapidly improve their performance with training; suggest how training needs might scale with task complexity; and demonstrate how advanced autonomous robotics techniques can help human operators improve their overall task performance. An additional contribution of this paper is to show how rigorous experimental and analytical methods from human factors research, can be applied to perform principled scientific evaluations of human test-subjects controlling robots to perform practical manipulative tasks. Gripper CameraSample test rig 7 DOF Robot Test objects for stacking task

The ability to predict how objects behave during manipulation is an important problem. Models informed by mechanics are powerful, but are hard to tune. An alternative is to learn a model of the object's motion from data, to learn to predict. We study this for push manipulation. The paper starts by formulating a quasi-static prediction problem. We then pose the problem of learning to predict in two different frameworks: (i) regression and (ii) density estimation. Our architecture is modular: many simple, object specific, and context specific predictors are learned. We show empirically that such predictors outperform a rigid body dynamics engine tuned on the same data. We then extend the density estimation approach using a product of experts. This allows transfer of learned motion models to objects of novel shape, and to novel actions. With the right representation and learning method, these transferred models can match the prediction performance of a rigid body dynamics engine for novel objects or actions.

No abstract

No abstract

Abstract:We demonstrate the resonant-like behaviour of the cardiopulmonary system in healthy people occurring at the natural low frequency oscillations of 0.1 Hz, which are often visible in the continuous pressure waveform. These oscillations represent the spontaneous oscillatory activity of the vasomotor centre and are sometimes called the Mayer waves. These 10-second rhythms probably couple with forced breathing at the same frequency and cause the observed cardiopulmonary resonance phenomenon. We develop a new method to study this phenomenon, namely the averaged Lomb-Scargle periodogram method, which is shown to be very effective in enhancing common frequencies in a group of different time series and suppressing those which vary between datasets. Using this method we show that in cardiopulmonary resonance the cardiopulmonary system behaves in a very similar way to a simple mechanical or electrical oscillator, i.e. becomes highly regular and its averaged spectrum exhibits a clear dominant peak and harmonics. If the forcing frequency is higher than 0.1 Hz, the total power and the share of power in the dominant peak and harmonics are lower and the prominence of the dominant peak and its harmonics greatly diminishes. It is shown that the power contributions from different forcing frequencies follow the resonance curve.