Quantifying Demonstration Quality for Robot Learning and Generalization

TL;DR

This paper addresses the problem of varying demonstration quality in Learning from Demonstration by proposing a task-performance–based quality metric. It fits a $K$-component $TP$-GMM with $GMR$ to demonstrations and uses a threshold $\delta$ to label demonstrations as high- or low-quality, then tests generalization on adjacent tasks with a $7\times7$ grid of targets. A two-session user study with $N=27$ participants (and $54$ demonstrations per session) reveals a strong correlation ($R=0.85$, $p<0.0001$) between task performance and generalization, and identifies fast-adapter vs slow-adapter user groups whose learning trajectories differ but converge with practice. The findings support using demonstration quality assessment to predict adaptation and learning efficiency, informing when and how demonstrations should be used for robot learning in real-world settings, and aligning results with motor skill learning literature.

Abstract

Learning from Demonstration (LfD) seeks to democratize robotics by enabling diverse end-users to teach robots to perform a task by providing demonstrations. However, most LfD techniques assume users provide optimal demonstrations. This is not always the case in real applications where users are likely to provide demonstrations of varying quality, that may change with expertise and other factors. Demonstration quality plays a crucial role in robot learning and generalization. Hence, it is important to quantify the quality of the provided demonstrations before using them for robot learning. In this paper, we propose quantifying the quality of the demonstrations based on how well they perform in the learned task. We hypothesize that task performance can give an indication of the generalization performance on similar tasks. The proposed approach is validated in a user study (N = 27). Users with different robotics expertise levels were recruited to teach a PR2 robot a generic task (pressing a button) under different task constraints. They taught the robot in two sessions on two different days to capture their teaching behaviour across sessions. The task performance was utilized to classify the provided demonstrations into high-quality and low-quality sets. The results show a significant Pearson correlation coefficient (R = 0.85, p < 0.0001) between the task performance and generalization performance across all participants. We also found that users clustered into two groups: Users who provided high-quality demonstrations from the first session, assigned to the fast-adapters group, and users who provided low-quality demonstrations in the first session and then improved with practice, assigned to the slow-adapters group. These results highlight the importance of quantifying demonstration quality, which can be indicative of the adaptation level of the user to the task.

Figures (7)

• Figure 1: Human demonstrators kinesthetically teach a robot to press a button on a box on (a) a low-constraint face close to the robot and (b) a high-constraint face such that the robot must be maneuvered around the box in a tight space.
• Figure 2: The proposed approach for assessing demonstrations using learned task performance in a task with two levels of constraints.
• Figure 3: Average success rate distribution of task performance of Session-1 on both task constraints.
• Figure 4: Learned task success rate for both slow- and fast- adapters in the two sessions of two levels of constraints task.
• Figure 5: Learned task success rate for both slow- and fast- adapters in all trials. Blue bars represents Session-1 trials and red bars represent Session-2 trials.