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

2025-05-12

AI Summary

Learning-based controllers deployed on physical systems face dual challenges of safety certification and computational efficiency, particularly on resource-constrained embedded platforms and under large-scale data regimes. This paper proposes CoLSafe, the first algorithm to jointly guarantee safety—via control barrier functions—and optimality—via online optimization—with sublinear computational complexity in data size. CoLSafe integrates lightweight online optimization with data-driven stability analysis, eliminating the need for high-fidelity system models or offline precomputation. Evaluated on a 7-DOF robotic manipulator, CoLSafe achieves a three-order-of-magnitude improvement in real-time performance over baseline methods while satisfying all safety constraints with 100% reliability and maintaining comparable task-level performance.

Technology Category

• Intelligent Robots: Learning & Optimization for ROB
• Natural Language Processing: Safety and Robustness
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

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.

Problem

• Ensuring safety in learning-based control systems
• Reducing computational cost for large datasets
• Achieving optimality with lightweight algorithms

Innovation

• Lightweight safe learning algorithm CoLSafe
• Sublinear computational complexity growth
• Guarantees for safety and optimality