Screw Geometry Meets Bandits: Incremental Acquisition of Demonstrations to Generate Manipulation Plans
Dibyendu Das, Aditya Patankar, Nilanjan Chakraborty, C. R. Ramakrishnan, I. V. Ramakrishnan
AI summary
Problem
Demonstration-based robot programming lacks a rigorous way to measure if a set of examples is sufficient and systematically requests additional ones, leaving the learning process opaque and inefficient.
Approach
The system uses screw-geometric motion planning to evaluate task coverage, then applies multi-armed bandit optimization to identify workspace regions with low success probability and incrementally requests targeted kinesthetic demonstrations until confidence thresholds are met.
Key results
- Concrete measure of demonstration sufficiency based on task-space coverage
- Bandit-driven sampling strategy to pinpoint demonstration locations
- User study shows non-experts can teach complex tasks with fewer than 10 demonstrations
- Simulation validation of demonstration distribution requirements
Why it matters
Empowers non-experts to efficiently program robots for complex manipulation by automating demonstration acquisition and guaranteeing plan feasibility.
Abstract
In this paper, we study the problem of method- ically obtaining a sufficient set of kinesthetic demonstrations, one at a time, such that a robot can be confident of its ability to perform a complex manipulation task in a given region of its workspace. Although programming by demonstration has been an active area of research, the problems of checking whether a set of demonstrations is sufficient and systemati- cally seeking additional demonstrations have remained open. We present an approach for the robot to incrementally and actively ask for new demonstration examples, one at a time, until the robot can assess with high confidence that it can perform the task successfully. Our approach uses (i) a screw geometric representation of motion to generate manipulation plans from demonstrations, which makes the sufficiency of a set of demonstrations measurable; (ii) a sampling strategy based on PAC-learning from multi-armed bandit optimization to evaluate the robot’s ability to generate manipulation plans in a subregion of its task space; and (iii) a heuristic to seek additional demonstration from areas of weakness. We present results of a user study conducted with 22 participants (without any background in robotics) on two example manipulation tasks, namely pouring and scooping, to assess the utility and usability of our approach. The results show that a handful of examples (fewer than 10) were needed to successfully teach the robot to plan tasks. A short video supplement is available on YouTube: https://youtu.be/KbAPgIouIvo