How should robots be programmed in order to facilitate the acquisition of a specific motor skill? We suggest that the answer may reside in the inherent geometric and dynamic properties of the task. We first define a general framework for the description of a motor skill at different levels (body configuration, body end effectors, tool, and results), and discuss the role of redundancy in learning a novel motor skill. We review different forms of assistance that have been proposed in the literature in different contexts. We then suggest that the problem of designing an 'optimal' scheme of assistance for a particular motor task is related to that of formulating an 'optimal' controller for that task. Finally, we discuss how assistance should be regulated in order to prevent the 'slacking' effect and to maximize subjects' participation.

Robot assistance can facilitate motor learning by way of haptic demonstration of the correct motor pattern. This is the good of the guidance hypothesis. However, correct haptic guidance may reduce the activity of the learner, including the correction of errors which is thought to be important for the acquisition of certain motor tasks. This is the bad of the guidance hypothesis. Overall, the net effect of haptic guidance can be positive or negative, depending on the kind of motor-learning problem, the type of guidance, and the scheduling of guided and unguided movements. We explored means to enhance the benefits of robot assistance with a focus on the learning of novel visuo-motor transformations.

While a plethora of anecdotal evidence supports the value of haptic interaction in encouraging motor skill acquisition (e.g. in neurorehabilitation performed with a physiotherapist) and task completion (e.g. aircraft steering performed by the mechanically coupled pilots), very few studies have explored the mechanisms of these interactions in depth. This talk presents an experiment we have designed to systematically investigate the strategies used by a dyad of healthy subjects to perform a movement and attenuate disturbances. To succeed in this redundant task, the subjects can increase impedance by working in opposition, or by co-contracting. We found that the subjects converged to a few dyad-specific motor strategies, and analysed the reasons for this specialisation.

This presentaiton will show how the latest scientific findings on motor skill learning can be used to optimize robot-therapy in improved neuro-rehabilitation treatments. Motivation is an important factor in rehabilitation and is frequently used as a determinant of rehabilitation outcome; hence, the continuous challenging and assisting can yield substantial advantages in motor learning and improve motor coordination. To maximize the efficiency of the learning process in terms of its specified target outcome, the robot trainer needs to continuously regulate assistance corresponding to the observed performance. In this presentation we will give an overview of the state of the art of the performance-based training algorithms and show how the assistance provided by the robot can be automatically adapted to patients' disability. Details will be provided about how the appropriate performance indicators, based on measures calculated by the robot, can be derived for each specific learning task. Finally, some models will be presented to show how these robot measures can be used to identify the different time patterns of recovery.

Repeated practice is a common approach to promote the restoration of motor skills in humans with disability. A robot could be a suitable assistant for the repeated practice, but it must take into account the specifics of the biological systems that are compromised. In order to optimise the practice, it is important to provide the naïve person with effective support within a rich environment in a manner that is safe, yet challenging. Here we describe a "learn from an expert" scenario in which the robot trainer has learning capabilities, and in particular it is capable of building its internal representation of the motor task by mimicking an expert human. An expert human performs many repetitions of the same motor task while being connected to the robot. The robot generates no interaction or assistive forces, and serves the sole purpose of recording all the aspects of the movement. The outcome of the learning process is a stochastic model of the skilled gestures created by the "biological" teacher, and the underlying control law. The control law is used by the robot for the training of the naïve person (patient). In order to increase the patient's level of motivation, we integrated the practice into a Wii game. This integration required the development of a special electromechanical device between the robot trainee, trainer and the Wii console. We tested a scenario where the expert played the Wii game. The expert's movements and other performance-relevant information were recorded and used to set up a control of the assistance needed. The robot then provided physical assistance to the naïve person (patient).

While interacting with robots several motor brain modes are present and different brain activity corresponds to each of them. For example, passive and active movements and intention to move while being moved passively are some of the brain modes commonly involved in motor rehabilitation. In this talk we will start with an introduction to the state of the art of non-invasive brain computer interfaces for motor rehabilitation and to neural correlates of motor related brain modes. We will present electroencephalography (EEG) data recorded while healthy subjects and stroke patients were involved in several motor related brain modes commonly present in motor rehabilitation therapies and some results of the use of a develop online haptic-BCI.