Deliberate Practice: Learning Robot Skills under a Budget

2026-08-13

AI Summary

This work addresses the critical challenge of efficiently acquiring skills for sequential tasks under a limited practice budget—a key bottleneck in robotic self-supervised learning. We propose the "Deliberate Practice" algorithm, which, for the first time, formulates skill acquisition under budget constraints as a bilinear programming problem that jointly optimizes estimates of skill mastery times, cumulative task rewards, and practice allocation. This unified optimization yields an exact solution for the budget-optimal practice strategy. Experimental results demonstrate that our approach significantly improves both practice efficiency and long-horizon planning performance in simulated and real-world manipulation tasks, achieving optimal utilization of scarce training resources.

Abstract

We consider the problem of autonomously learning robot skills under a limited practice budget for sequential tasks. We propose an active skill learning algorithm, \emph{Deliberate Practice (DP)}, that computes a provably \emph{budget-optimal} allocation---practicing skills that maximize expected cumulative reward while being learnable within the budget. DP estimates both the time needed to master skills and the cumulative reward of the task plans that the skills unlock. Computing a budget-optimal allocation is challenging as it requires reasoning about combinatorially many skill plans over a large practice budget. Our key contribution is a bilinear program that can compute this exactly using off-the-shelf solvers. Through simulated and real-world experiments on long-horizon manipulation tasks, we show that our approach allows robots to optimally use limited practice time to acquire useful policies and improve long-horizon planning.

Problem

• Deliberate Practice
• robot skill learning
• practice budget
• sequential tasks
• budget-optimal allocation

Innovation

• Deliberate Practice
• budget-optimal allocation
• bilinear programming
• active skill learning
• robotic manipulation