HARMO
25131
17/09/2026
From 2026 to 2029
HARMO develops and validates automated vehicles that take better account of how human driving behaviour changes with experience and familiarity with a route. TML leads the behavioural modelling and scenario-based validation, and translates the results into measurable safety and acceptance criteria.
Today, automated vehicles are often developed and tested using models in which human drivers behave in a relatively static manner. In reality, driving behaviour changes when drivers are familiar with a route: they allocate their attention differently, react more out of habit and may apply different safety margins. HARMO builds on the European projects i4Driving, BERTHA, HEADSTART, SUNRISE, and SYNERGIES, and explicitly incorporates these effects of learning, habituation and route familiarity into models of human driving behaviour. To this end, the project collects data from driving simulators and journeys in real traffic, and translates this into behavioural models that account for differences between familiar and unfamiliar situations. These models are used in simulations in which human and automated vehicles actually interact with one another. HARMO is also developing an AI planner that combines natural, human driving behaviour with explicit traffic and safety rules. Finally, realistic test scenarios, safety and acceptance criteria are being developed, and the models are being tested in real traffic in ‘shadow mode’: the system predicts what the vehicle and other road users would do, without taking control itself.
Today, automated vehicles are often developed and tested using models in which human drivers behave in a relatively static manner. In reality, driving behaviour changes when drivers are familiar with a route: they allocate their attention differently, react more out of habit and may apply different safety margins. HARMO builds on the European projects i4Driving, BERTHA, HEADSTART, SUNRISE, and SYNERGIES, and explicitly incorporates these effects of learning, habituation and route familiarity into models of human driving behaviour. To this end, the project collects data from driving simulators and journeys in real traffic, and translates this into behavioural models that account for differences between familiar and unfamiliar situations. These models are used in simulations in which human and automated vehicles actually interact with one another. HARMO is also developing an AI planner that combines natural, human driving behaviour with explicit traffic and safety rules. Finally, realistic test scenarios, safety and acceptance criteria are being developed, and the models are being tested in real traffic in ‘shadow mode’: the system predicts what the vehicle and other road users would do, without taking control itself.
Within HARMO, TML is leading the research into learning and route familiarity in human driving behaviour (WP1) and into the scenario-based verification and validation of automated vehicles (WP3). We translate experimental driving data into parameters for behavioural models, define realistic traffic scenarios, build closed-loop simulations using human behavioural models, and develop measurable KPIs and assessment criteria for safety, predictability, and human driving behaviour. In addition, TML is leading the development of the Acceptance Threshold Matrix, which determines how the acceptability of automated driving behaviour can be measured for drivers, passengers, and other road users.
