Context
Automated enforcement programs operate under limited resources and heterogeneous road conditions. The decision is not simply whether enforcement works on average, but where deployment can create the most value.
Decision challenge
Observed outcomes reflect the locations that were actually selected, so policy learning must be grounded in observational evidence rather than a clean randomized experiment. The allocation problem is also sequential: deployment choices affect which outcomes are observed next.
Approach
The project develops an observational policy-optimization framework with reinforcement learning. The aim is to connect empirical evidence with an explicit resource-allocation decision.
Why it matters
The project illustrates a broader research principle: predictive or causal evidence becomes operationally useful only after it is connected to the policy that must be chosen.