HARMO solves a critical problem in automated driving: Current systems struggle in cities because they cannot anticipate how local, "habituated" drivers behave. While most models assume drivers are always alert, real drivers on familiar routes often rely on relaxed, automatic habits. HARMO bridges this gap by delivering TRL 5 models that account for "familiarity," enabling automated vehicles to better predict and interact with the implicit, often less attentive behaviours typical of mixed traffic. HARMO aims to:
1. Model Familiarity: Quantify how repeated route exposure alters driver attention and risk acceptance.
2. Generate Human-Like Control: Develop a Neurosymbolic-Diffusion planner that combines the natural fluidity of generative AI with verifiable symbolic safety constraints.
3. Validate in Mixed Traffic: Operationalise these models in closed-loop simulations where background agents exhibit realistic learning curves.
4. Define Regulatory Criteria: Establish an "Acceptance Threshold Matrix" to certify predictability and trust.
5. Demonstrate in Real-World: Validate the framework via shadow-mode integration in urban pilots.

Building on results from i4Driving and BERTHA, HARMO treats familiarity as a dynamic latent variable. We generate a "Habituation Dataset" to calibrate HBMs, which are then integrated into a hybrid AI planning stack. This stack is validated using EU-CEM compliant scenarios, replacing static background traffic with "smart" agents. Finally, the system is benchmarked in real-world logistics and shuttle pilots, comparing HARMO’s behavioural predictions against live human traffic without altering vehicle control. By translating scientific models into "Safety Evidence Packs" for type approval, HARMO enables the design of "safe, human-like behaviour" that is predictable to other road users, supporting the trustworthy deployment of CCAM in Europe.