Safety and optimality in learning-based control at low computational cost

TL;DR

This paper introduces CoLSafe, a lightweight safe learning algorithm that ensures safety and near-optimality for physical systems, with low computational costs suitable for embedded devices and large datasets.

Contribution

The paper presents CoLSafe, a novel safe learning method with sublinear computational complexity and formal safety and optimality guarantees.

Findings

Successfully applied to a seven-degrees-of-freedom robot arm

Achieves safety and near-optimality with low computational cost

Demonstrates scalability to large datasets and real-time systems

Abstract

Applying machine learning methods to physical systems that are supposed to act in the real world requires providing safety guarantees. However, methods that include such guarantees often come at a high computational cost, making them inapplicable to large datasets and embedded devices with low computational power. In this paper, we propose CoLSafe, a computationally lightweight safe learning algorithm whose computational complexity grows sublinearly with the number of data points. We derive both safety and optimality guarantees and showcase the effectiveness of our algorithm on a seven-degrees-of-freedom robot arm.