Analyzing Therapeutic Interactions in Robot-Led Rehabilitation

The robot-led training sessions had audio-visually been video recorded twice, both on the first and the last (9th) session. Hence, for each participant, data of the two sessions were available for offline rating of the therapeutic interactions. The videorecorder was placed to cover the therapy scenario, its agents, the interface used for visual displays (tablet or monitor), and to show the training activity currently performed. Since the therapeutic interaction implemented in the system is either verbal or audio-visual accompanied by verbal phrases, the audio-recording was also mandatory and used for the analysis of therapeutic interactions.
The scenarios, as video recorded, differed for the following two types of trainings (compare Figure 1):

A. AAT: for stroke patients with mild arm paresis, the scenario with a patient, humanoid robot, and supervising staff (three interactive agents).

B. ABT: for patients with moderate-to-severe arm paresis, the scenario with a patient, humanoid robot, helper, and supervising staff (four interactive agents).

Even though the robot is programmed to provide all therapeutic interactions necessary, there might be situations where the therapist or the helper steps in naturally and spontaneously (they are not given instructions to do so) and provides additional therapeutic interactions.
Therefore, any therapeutic interaction as performed either by a robot, therapist, or helper was documented.
The two trained raters (Ann Louise Pedersen and Philipp Deutsch) independently analysed and documented the therapeutic interactions observed in the two video-recorded sessions per participant using the instrument THER-I-ACT and its manual. THER-I-ACT measures both the occurrence/frequency and the timing of the therapeutic interactions in the thematic fields of “information provision,” “feedback,” and “bonding” with a variety of pre-defined categories in each thematic field and in addition provides a global rating of the focussed attention and engagement for both the patient and therapist (for details, see Platz et al., 2021 (link)).