Safe & Constrained RL

CMDP, safe RL, constrained policy optimization, Lyapunov methods, feasibility, and safety guarantees.

All research notes
Safe & Constrained RL

CAffNet: Hard Constraint-Affine Neural Networks

CAffNet builds input-dependent affine constraints into a neural output layer. It guarantees constraint satisfaction when the supplied feasible set is nonempty. Enumerating active constraint combinations can be costly, and physical safety still depends on the accuracy of the constraint model.

CAffNet: Hard Constraint-Affine Neural Networks

Safe & Constrained RL

The Safety–Information Deadlock in Adaptive Probabilistic Shielding

How online MDP estimation, interval uncertainty, and probabilistic shielding form a coupled loop in which blocking risky actions can also block the evidence needed to learn safety.

Safe & Constrained RL

Hard Linear Constraints in Neural Decisions: Feasibility by Decision-Rule Anchoring

A critical note on enforcing input-dependent linear equality and inequality constraints in neural network outputs using a robustly feasible decision-rule anchor and minimal interpolation.

Enforcing hard linear constraints in deep learning models with decision rules

Safe & Constrained RL

Lyapunov Projection for Continuous-Action Safe Reinforcement Learning

This note reads Lyapunov-based safe policy optimization as a practical projection bridge from finite CMDP safe policy iteration to continuous-action deep reinforcement learning, while separating exact CMDP guarantees from local approximation behavior.

Lyapunov-based Safe Policy Optimization for Continuous Control

Safe & Constrained RL

Lyapunov-Based Safe Policy Improvement for CMDPs

Lyapunov constraints turn an expected cumulative safety budget in a CMDP into local restrictions on policy improvement. This note examines the exact certificate logic and the weaker status of neural approximations.

A lyapunov-based approach to safe reinforcement learning