Enabling Robots to Identify Missing Steps in Robot Tasks for Guided Learning from Demonstration

Abstract: Learning from Demonstration (LfD) systems are commonly used to teach robots new tasks by generating a set of skills from user-provided demonstrations. These skills can then be sequenced by planning algorithms to execute complex tasks. However, LfD systems typically require a full demonstration of the entire task, even when parts of it are already known to the robot. This limitation comes from the system's inability to recognize which sub-tasks are already familiar, leading to a repetitive and burdensome demonstration process for users. In this paper, we introduce a new method for guided demonstrations that reduces this burden, by helping the robot to identify which parts of the task it already knows, considering the overall task goal and the robot's existing skills. In particular, through a combinatorial search, the method finds the smallest necessary change in the initial task conditions that allows the robot to solve the task with its current knowledge. This state is referred to as the excuse state. The human demonstrator is then only required to teach how to reach the excuse state (missing sub-task), rather than demonstrating the entire task. Empirical results and a pilot user study show that our method reduces demonstration time by 61% and decreases the size of demonstrations by 72%.

• The paper introduces a novel approach where robots generate 'excuses' to identify minimal changes needed for task success.
• It utilizes an automated excuse generation algorithm in VR, achieving a 54% reduction in demonstration time and a 74% reduction in demo size.
• This method streamlines demonstrations by focusing human guidance on specific missing skills rather than full task execution.

Understanding Robots through "Excuses"

The Challenge of Robot Learning

Teaching robots to perform tasks by demonstrating actions to them (known as "Learning from Demonstration" or LfD) is a prevalent method to enable robots to interact with their environment. Despite the promise of this approach, robots often face limitations in understanding which parts of a task they can accomplish based on their existing knowledge. Traditional systems require complete task demonstrations, even for actions that fall within the robot's current capabilities—resulting in inefficiency. This means a human demonstrator, even for missing skills in a task, has to unnecessarily demonstrate the entire task. This process can be both time-consuming and repetitive, potentially reducing a human's willingness to engage with the robot.

Excuses to the Rescue

The concept of "excuses" is introduced as a solutions-based approach where a robot identifies the smallest change in its state that would render an otherwise unsolvable task achievable. An "excuse" in this context implies what a robot would need to learn or change to complete a given task. For instance, if a robot required to clean a kitchen lacks the skill to open drawers, it could signal an excuse like, "if only the drawer was open, I could complete my task." This would allow a human demonstrator to show the robot specifically how to open the drawer, instead of having to provide a full task demonstration. Thus, enabling more focused and useful learning sessions for the robot.

Empirical Validation

To validate the usefulness of excuses, a study was conducted using a Virtual Reality environment and an automated excuse generation algorithm. In a user study with various task scenarios, the use of excuse-based demonstrations led to a considerable reduction in both the time taken and the size of necessary demonstrations needed for the robot to perform the tasks independently. The use of excuses resulted in a 54% reduction in demonstration time and a 74% reduction in demonstration size, marking a significant improvement in the effectiveness of LfD methodologies to expand robot capabilities.

Looking Ahead

The integration of excuses into the LfD framework offers a promising direction for robotic task learning. Not only do excuses streamline the demonstration process by identifying what is precisely needed for task completion, but they also spare humans from demonstrating familiar actions. However, future work is necessary to enhance guidance from excuses, as they may vary in size and impact on the demonstration. Additionally, understanding user preferences and improving the delivery of excuses to demonstrators remain areas ripe for further research.

The evolution of LfD with excuses opens up avenues for better human-robot collaboration, potentially easing the transition of robots into roles where they can assist effectively in dynamic, real-world environments.