Quantifying Demonstration Quality for Robot Learning and Generalization

Sakr, Maram; Li, Zexi Jesse; Kulić, Dana; Croft, Elizabeth A.

doi:10.1109/lra.2022.3191950

Cited by 6 publications

(6 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Others changed their strategies from dense point sets to a more distributed approach, albeit with scepticism. These findings suggest the potential need for adaptive training tailored to individual participants' requirements, especially for those with strong pre-training strategy convictions, aligning with previous research [9].…”

Section: Discussionsupporting

confidence: 85%

See 1 more Smart Citation

What Makes a Good Demonstration for Robot Learning Generalization?

Sakr

Kulić

Croft

2021

Companion of the 2021 ACM/IEEE International Conference on Human-Robot Interaction

View full text Add to dashboard Cite

Learning from Demonstration (LfD) is a framework that allows lay users to easily program robots. However, the efficiency of robot learning and the robot's ability to generalize to task variations hinges upon the quality and quantity of the provided demonstrations. Our objective is to guide human teachers to furnish more effective demonstrations, thus facilitating efficient robot learning. To achieve this, we propose to use a measure of uncertainty, namely task-related information entropy, as a criterion for suggesting informative demonstration examples to human teachers to improve their teaching skills. In a conducted experiment (N=24), an augmented reality (AR)based guidance system was employed to train novice users to produce additional demonstrations from areas with the highest entropy within the workspace. These novice users were trained for a few trials to teach the robot a generalizable task using a limited number of demonstrations. Subsequently, the users' performance after training was assessed first on the same task (retention) and then on a novel task (transfer) without guidance. The results indicated a substantial improvement in robot learning efficiency from the teacher's demonstrations, with an improvement of up to 198% observed on the novel task. Furthermore, the proposed approach was compared to a state-of-the-art heuristic rule and found to improve robot learning efficiency by 210% compared to the heuristic rule.

show abstract

Section: Discussionsupporting

confidence: 85%

“…Teaching robots through demonstration poses challenges for lay users as they may struggle to provide useful [9], [7] and sufficient demonstrations [8]. Robots can become more proactive in their learning by requesting additional demonstrations [10] or seeking clarification [11].…”

Section: Related Workmentioning

confidence: 99%

What Makes a Good Demonstration for Robot Learning Generalization?

Sakr

Kulić

Croft

2021

Companion of the 2021 ACM/IEEE International Conference on Human-Robot Interaction

View full text Add to dashboard Cite

show abstract

“…This can result in incorrect policies learned. As mentioned above, the existing solution focuses on improving the performance of the learner by filtering the demonstration to remove misleading examples to improve the quality of the learned policy Sakr et al (2022) Mahler and Goldberg (2017).…”

Section: Goals and Research Questionsmentioning

confidence: 99%

“…While technological advances have significantly improved robotic performance, however, the robot's learned behavior becomes a black box, making it difficult for non-expert demonstrators to discern how and what it has learned during the teaching process. In this context, researchers propose developing a more sophisticated model to handle ambiguous demonstrations Sena et al (2019); Sakr et al (2022); however, an alternative solution is to take advantage of the natural adaptability of the human demonstrator to improve the performance of the system.…”

Section: Introductionmentioning

confidence: 99%

“…Since the efficacy of the learned policy depends on the quality of interactions and data provided by humans Sakr et al (2022), it is important to assess how communicating the robot's learning to nonexpert teachers can impact their behavior specifically in terms of their ability to teach and handle the workload. Therefore, to determine the impact of communicating the robot's learning to non-expert demonstrators, we propose an empirical study that facilitates the formal assessment of human performance based on robot learning knowledge, considering state-of-the-art methods.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Understanding and Quantifying Human Factors in Programming from Demonstration: A User Study Proposal

Mehak,

Jain,

Kelleher

et al. 2023

Proceeding of the 33rd European Safety and Reliability Conference

View full text Add to dashboard Cite

Programming by demonstration (PbD) is a promising method for robots to learn from direct, non-expert human interaction. This approach enables the interactive transfer of human skills to the robot. As the non-expert user is at the center of PbD, the efficacy of the learned skill is largely dependent on the demonstrations provided. Although PbD methods have been extensively developed and validated in the field of robotics, there has been inadequate confirmation of their effectiveness from the perspective of human teachability. To address this gap, we propose to experimentally investigate the impact of communicating robot learning process on the efficacy of the transferred skills. This paper outlines the preliminary steps in designing experiments to identify human-related performance shaping factors in PbD. The purpose of this article is to establish the foundation for an experimental study that focuses on the human component in PbD algorithms and provides new insights into human factors in PbD design.

show abstract