In a paper published in the journal PLOS One, researchers delved into end-user robot programming, specifically examining kinesthetic teaching, which allowed non-experts to guide robots rather than write code physically. Despite its promise, this approach posed challenges due to users' unfamiliarity with kinematics and limitations in robot capabilities.
Study: Enhancing End-User Robot Programming: A Comparative Study of Kinesthetic Teaching Approaches. Image credit: Summit Art Creations/Shutterstock
The study compared self-guided practice with curriculum-based training and surprisingly found no significant difference in programming proficiency. However, it uncovered insights into factors affecting programmers' success, highlighting the need for refined learning interventions in end-user robot programming. This research served as a launchpad for devising more effective training methods to enhance programming skills in this domain.
Enhancing Robot Programming
End-user robot programming methods, like kinesthetic teaching, allow manual guidance of robots instead of coding, facilitating the customization of collaborative robots' behaviors. While kinesthetic teaching offers accessibility and reduced cognitive load, users face challenges in creating practical demonstrations due to unfamiliarity with robot capabilities and physical demands.
Traditional solutions focus on algorithms and interfaces, but a skills gap persists, prompting recent efforts toward educational tools. These interventions aim to enhance users' programming skills, influencing their perceptions of robots. Historically, robot programming demanded extensive training, but with cobots, shorter learning periods suffice through classes, simulations, and minimal training options. However, kinesthetic teaching's motion-level education needs more structure, relying heavily on unguided practice. Curriculum-based training, known for structured learning sequences, emerges as a potential solution.
Assessment of Kinesthetic Teaching Outcomes
The research team conducted a user evaluation to examine how practice-based and curriculum-based learning affected programming proficiency and user perceptions. The study comprised two one-hour sessions, where participants engaged in kinesthetic teaching tasks using a UR5 robot arm and teach pendant. The first session varied depending on the assigned study condition—either practice or curriculum. For the practice group, participants focused on programming the Universal Robots Version 5 (UR5) to build a tower using toy blocks, simulating pick-and-place practice tasks typical in kinesthetic teaching. Conversely, the curriculum group followed a designed curriculum.
In the second session, standard across both conditions, participants undertook four target tasks to assess their learning outcomes. These tasks—insertion, pouring, hanging, and stacking—replicated actions commonly performed by collaborative robots in real-world contexts. The tasks assessed participants' transfer abilities by testing their replication of practiced skills in different contexts. Additionally, they evaluated participants' far transfer skills by assessing their capability to generalize skills to varied manipulation tasks.
Researchers designed the curriculum design by learning science and empirical exploration. It comprised three modules focusing on essential component skills for effective kinesthetic teaching: individual joint motion, gripping, and planning. Each module integrated hands-on activities and proficiency-based learning, allowing learners ample time to master each skill before progressing. The curriculum employed a bottom-up learning approach, emphasizing mastering component skills and their integration into task-oriented contexts. Additionally, it utilized strategies such as contrast-based teaching and variation patterns to enhance users' understanding of joint motion and its relation to end effector positions.
The hypotheses centered on learners following the curriculum exhibiting superior programming confidence and proficiency compared to self-guided practice. To gauge user experience, researchers collected data on changes in confidence levels related to UR5 programming, planning, and executing kinesthetic teaching, along with projected confidence in unfamiliar tasks and perceived workload, using adapted National Aeronautics and Space Administration Task Load Index (NASA TLX) scales.
Researchers measured task success by tallying the number of completed tasks and unsuccessful demonstrations, while they assessed task efficiency by tracking the time taken to complete each task. Program quality assessments considered program suboptimality and the average rates of change of force and torque. Additionally, user gaze behaviors were recorded to explore differences in fixation duration and quantity, especially on the robot gripper, indicative of skill levels in kinesthetic teaching. This comprehensive approach allowed the gathering of multifaceted data on programming proficiency and user perceptions.
Kinesthetic Teaching: Learning and Evaluation
The study underwent rigorous ethical approval and consisted of two sessions separated by approximately 24 hours. During the initial session, the facilitators introduced participants to the essential tools and procedures for kinesthetic teaching using the UR5 robot arm. They engaged in practice-based or curriculum-based learning tasks within a 40-minute timeframe, following a familiarization task and instructions on using the teach pendant.
The subsequent session involved participants undertaking four targeted tasks to evaluate their acquired skills, aiming to assess both near and far transfer abilities. Minimal experimenter intervention ensured independent participant operation during these tasks. Post-task completion, participants completed a questionnaire and participated in an interview to provide insights into their learning experiences.
The study recruited 28 diverse participants, primarily from engineering and technology backgrounds, and noted one incomplete participation from the practice group. Participants varied in age and prior experience with technology, programming, and robots, but none had previous exposure to kinesthetic teaching. Researchers used multiple metrics to evaluate user experience, task success, efficiency, program quality, and gaze behaviors. However, user experience, task success, progress, and efficiency metrics were similar between practice-based and curriculum-based learners.
While participants encountered varied challenges and successes in different tasks, both groups showed comparable completion rates and task efficiency. The qualitative insights from participant interviews provided valuable context, showcasing factors contributing to success, such as repeated interactions with the robot and the importance of trial and error in building confidence and proficiency in programming.
Conclusion
To sum up, the study showed no significant differences between open-ended practice and the designed curriculum for kinesthetic teaching. Nonetheless, it emphasized the pivotal role of initial user experiences in shaping programming confidence and perceptions. Adding structure to training interventions could enhance user experiences and performance, yet the complexity of learning programming warrants further exploration of various teaching factors for better accessibility to non-experts in programming methods like kinesthetic teaching.
Advancing Large Language Models with Multi-Token Prediction