This section will explore the human-robot interaction taxonomy data (Figures
11 and
12) as informed by human-robot interaction and robot classification taxonomies (e.g., [
40,
464]). Detailed explanation of taxonomy hierarchy and their relevant classification labels can be found elsewhere [
40,
464], and a brief summary of labels has been listed in the review information and categorization section (Section
3.1). Task type was relatively broad with limited consistency between studies and has been reported in individual sections listed prior to this section (see Section
6). Task criticality for most robot use cases was identified as low criticality (
N = 271, 88%) compared to medium (
N = 32, 10%) or high (
\(N=7\), 2%) classification, which may further support the emergent nature of robot roles in easier use cases as a first application. There was also a link between task difficulty and frequency, where less difficult tasks are more commonly investigated and harder tasks are less represented. These classification patterns were similar to our own custom metric on a single score for overall task evaluation using task complexity, risk, importance, and robot complexity: low (
N = 249, 80%), medium (
N = 51, 17%), and high (
\(N=10\), 3%). Robot morphology was categorized as anthropomorphic (human-like), zoomorphic (animal-like), and functional (neither human-like nor animal-like, but related to function). Most systems were functional (
N = 203, 65%) compared to anthropomorphic (
N = 104, 34%) or zoomorphic (
\(N=3\), 1%). Functional robots were more likely to be used in medium and high task criticality studies compared to anthropomorphic or zoomorphic robots. A high volume of works had a 1:1 human to robot ratio (
N = 281, 91%) with an overall mean of 1.08, 9 with more than one robot, 20 with more than one human, a maximum reported ratio as 5 [
246], and minimum reported ratio between 0.1 [
13] and 0.07 [
318]. Human-robot teams often had 1:1 human-robot team compositions, showing that the methods and team setups focused on a single human to potentially assist in robustness and utility of robotic vision in the collaborative scenario. Homogeneous teams were used for 307 (99%) of the reported studies, with only 3 (1%) of studies using heterogeneous robot team compositions (e.g., [
86,
238,
253]). Level of shared interaction among teams was high in the ‘A’ formation (
N = 280, 90%) as predicated on the earlier reported ratio of people to robots (
N = 281, 90%). There were 8 papers with a ‘B’ formation (one human with multiple robots using a single interaction), 5 with a ‘C’ formation (one human with multiple robots using a separate interaction for each robot), 6 with a ‘D’ formation (multiple humans with one robot, where the robot interacts with the humans through a single interaction), and 11 with an ‘E’ formation (multiple humans with one robot, where each human interacts with the robot separately). In terms of the type of human-robot physical proximity, interacting (
N = 186, 60%) was the highest followed by following (
N = 64, 21%) and then avoiding (
N = 24, 8%). No studies that used tele-operation were eligible in this review, but for the remainder of eligible studies, nearly all (
N = 309, 99%) had the robot as synchronous (same time) and collocated (same place) with the exception of one study that was non-collocated (e.g., [
229]). Autonomy level scoring by Beer et al. [
40] was used, but no scores were classified on level 1 due to exclusion criteria that the robot must not be manually operated by the human. For the remainder, robots were often high on autonomy, which may have been skewed by initial entry criteria that required the robot to use robotic vision to perform an action or response. Figure
13(a) depicts that papers over the past 10 years often had level 2 (tele-operation: robot prompted to assist but sensing and planning left to the human), level 6 (shared control with human initiative: robot senses the environment, develops plans/goals, and implements actions while the human monitors the robot’s progress), or level 10 (full autonomy: robot performs all task aspects autonomously without human intervention). Figure
13(b) depicts that there was a relatively even spread of autonomy level across the four domains, Figure
13(c) depicts mobile and fixed manipulators were most often used with level 10 autonomy, with similar trends seen across the autonomy levels, and Figure
13(d) depicts camera types per robot autonomy level.