RAMPS project

RAMPS: Robust Adaptive Multi-Step Predictive Shielding — Deep Dive

RAMPS combines learned linear dynamics with multi-step robust control barrier functions to make safe reinforcement learning practical in complex, high-dimensional environments.

What this article covers

This article explains the safety problem, model and shielding method, experimental results, implementation considerations, assumptions, and limitations. For formal proofs and code, use the primary sources below.

Read the paper · View code