Stroke is a primary cause of long-term disability worldwide  with nearly 1.1 million persons in Europe suffering a stroke each year . Importantly, this number is expected to increase to more than 1.5 million cases per year in 2025, mainly due to an aging population .
Approximately 70–85% of persons post-stroke present with impairment of an upper limb [4, 5] that persists even after 3–6 months from stroke , leading to a significant reduction of independence and quality of life . Consequently, improving upper limb functionality is a core element of stroke rehabilitation to reduce disability and increase the capacity to perform the activities of daily living (ADLs) . Different rehabilitative approaches have been proposed [9, 10], including constraint induced movement therapy , functional electrical stimulation [12, 13], virtual reality [14, 15] and robot therapy [16, 17]. Regarding the latter approach, two recent reviews [16, 17] indicated that robot-based rehabilitation is effective in improving ADLs, arm function and muscle strength in persons post-stroke. Previous studies suggested that the advantage of robotic devices, when compared with other physiotherapy approaches, may be the capability of these systems to provide rehabilitation paradigms enabling a strict application of some motor learning principles [18,19,20] indispensible to promote neural plasticity and reorganization [21,22,23]. In particular these principles include (1) the provision of highly intensive training involving a large number of goal-directed movements (e.g. center-out reaching of peripheral targets aimed at improving the coordination between shoulder and elbow) [21, 24], (2) the promotion of active participation by the person, also when severely impaired , and (3) the provision of real-time sensory feedback (visual and haptic) and quantitative summary feedback that can be used by the participant to correct his/her movement [14, 26]. Importantly, as previously discussed [27, 28], further investigation is needed to evaluate if the application of these motor learning principles can enhance the transfer of the rehabilitation effects also to non-trained tasks and contexts typical of ADLs.
The effects of motor rehabilitation on upper limb function are commonly assessed with clinical scales  that are mainly focused on task accomplishment, but do not give quantitative, objective and sensitive information on underlying changes in neuromotor control strategies involving inter-joint coordination and/or compensatory movements [30,31,32,33]. As discussed by Levin et al. , the main goal of motor rehabilitation is to lead the person to accomplish a task. However, also the assessment of how the task is performed is of paramount importance to evaluate whether the person has regain the ability to execute the task with a more physiological upper limb motor pattern (recovery), or he/she has developed compensatory strategies, such as abnormal trunk rotations (compensation) [30, 31, 34,35,36,37]. Instrumented motion analysis may provide this information and complement clinical assessment [31,32,33, 38, 39].
Instrumented analysis is usually performed using quantitative robot-based indexes describing a number of trained and non-trained tasks [28, 40,41,42,43,44]. As summarized in a review by Nordin et al. , the most common robot-based parameters describing upper limb movement and sensation include the amplitude of robot-generated forces [40, 41], temporal and speed metrics [40, 43, 44, 46, 47], response latency [46, 47], accuracy indexes [40, 43, 44, 46, 47], path length and range of motion [41, 42, 46, 47], and movement smoothness [40,41,42,43,44, 46, 47]. The test-retest reliability, the discriminant ability and the concurrent validity of these robot-based indexes have been analyzed in a large number of studies. Among these studies, those including the largest samples of persons post-stroke [41, 46,47,48,49] found good to excellent reliability [41, 48], good discriminant ability [41, 47], and moderate to high concurrent validity with clinical scales [41, 46, 47, 49]. The main advantage of the robot-based indexes is that they can be easily obtained during the course of the robotic training, thus providing indications about the gradual progression of the participants’ performance . By contrast, the main drawback is that these parameters mainly describe the trajectory of the end-effector during planar tasks executed within the robot workspace that is different from the typical daily living contexts.
This drawback may be partly overcome by using more sophisticated kinematic analysis techniques [32, 33, 38, 51,52,53,54,55,56,57] aimed at characterizing the execution of more ecological activities performed outside the robot workspace, including pointing tasks [34, 37] or reaching forward and touching real objects placed on a table, such as boxes [54, 55], cups , glasses [32, 33, 57], discs , cones  and desk bells [52, 53, 56]. Compared to the robot-based indexes, these analyses may provide a more detailed characterization of the different components of a task (e.g. upper limb and trunk movements), thus adding information about the way a task is performed before and after a rehabilitation treatment. This, in turn, may help in assessing the effects of such treatment in terms of neuromotor recovery and/or compensation [30, 34, 37, 50]. However, with the exception of Cirstea and Levin  who described trunk and arm motion during a 3D pointing tasks, all the above mentioned studies analyzed activities that mainly involved movements in the horizontal plane, with a minimal vertical component against gravity that is, however, a fundamental aspect of ADLs.
Following these considerations, this pilot study had two aims. The first aim was to assess the effects of planar robotic rehabilitation versus arm-specific physiotherapy in persons post-stroke on motor strategies derived from instrumented kinematic analysis of upper limb and trunk during the execution of a non-trained task involving horizontal and vertical arm movements. The second aim was to compare the effects of the two rehabilitation approaches on arm function as measured by clinical scales. We hypothesized that robot therapy provides larger improvements in the coordination between shoulder and elbow joints and in upper limb impairment, since it enables a rigorous application of the motor learning principles described above, in particular administration of high intensity goal-directed training, promotion of active participation, and provision of feedback.