Using PDDL Planning to Ensure Safety in LLM-based Agents

Integrates PDDL symbolic planning with LLM-based agents to enforce safety constraints during execution. Experiments show robustness under severe input perturbations and adversarial attacks.